refactor(infra-local): simplify + unify into one Python compose package (~50% smaller) (#796)

## What

Simplifies `apps/infra-local` (the BoxLite-based local dev stack) and
unifies its two-language orchestrator into one Python package. **~4,500
→ ~2,190 lines (~50%)**, one entry point (`python -m compose`), flat
structure.

## Commits

1. **simplify** — drop dead config (the `BOXLITE_*_HOST_PORT` knobs
never moved a bound port; `pg_url`/`tcp_port`/`Severity.WARN`; dead
`configs/minio/init.sh`), dedupe the SDK import shim, delete the stale
`CONNECTIONS.md` (445 lines restating the code), rewrite the README, and
remove the infra-local test suite (not in CI; mostly asserted dataclass
defaults) — preserving the one unique capability as a native-gated SDK
test.
2. **port L2 to Python + rename + flatten** — the 4 native host
processes (API/runner/proxy/dashboard) move from ~800 lines of bash
(`scripts/`) into `compose/native.py` (subprocess supervision: detached
daemons, pidfiles, `killpg` teardown). Package `boxlite_local` →
`compose`. `scripts/` + `configs/` deleted; `api.env` → root. CLI is 7
verbs (`up/down/status/logs/restart/reset/nuke`); Makefile ~19 → 9 thin
aliases.
3. **adversarial fixes** — a 2-round adversarial-review loop found +
fixed an `export KEY=val` parser bug in the `.env` reader; the SDK
capability test was rewritten (`alpine` busybox lacks `httpd`) to a
`python:3-alpine` http.server box and verified live.

## Final structure

4 root files (`api.env`, `Makefile`, `README.md`, `pyproject.toml`) +
the `compose/` package (config / services / orchestrator [L1] / native
[L2] / doctor / _sdk / __main__). No `scripts/`, `configs/`, or
`tests/`.

## Verification

- `python -m compose --help` + all 7 subcommands; `py_compile` + import
all modules.
- SDK volume-port test passes **live** (`pytest -m integration` → 1
passed, 30s): RW-volume persistence + host-port mapping.
- 2-round adversarial review terminated at `ZERO_FINDINGS`; CLAUDE.md
auditor PASS on every commit; pre-push integration suite green.

## ⚠️ Reviewer note — live-stack validation needed before merge

`compose/native.py` is a faithful behavior-port of the deleted bash, but
**process supervision is runtime-only behavior**. It
compiles/imports/parses and the adversarial review cleared it against
the bash spec, but a live **`make up && make status`** (then `make down`
/ `make restart COMPONENTS=runner`) on an Apple Silicon Mac is the final
proof the L2 daemons spawn/reap correctly. Please run that before
merging.

No Cargo/Rust/core-SDK source changes; scope is `apps/infra-local/` +
one `sdks/python/tests/` file.

<!-- This is an auto-generated comment: release notes by coderabbit.ai
-->
## Summary by CodeRabbit

# Release Notes

* **Refactor**
* Simplified local infrastructure orchestration: replaced the legacy
`boxlite_local` workflow with `python -m compose`; Makefile now exposes
thin aliases for `up`, `down`, `status`, `logs`, `restart`, `reset`, and
`nuke`.
* Simplified configuration and standardized the persistent state
directory to `.apps-local/`.

* **Documentation**
* Updated `apps/infra-local` README for the new `compose` workflow and
endpoints.
  * Removed legacy connection-reference documentation.

* **Tests**
* Added a Python SDK integration test covering RW volume persistence and
port forwarding across restarts.
  * Removed legacy infra-local integration/unit test suites.
<!-- end of auto-generated comment: release notes by coderabbit.ai -->

---------

Signed-off-by: dorianzheng <8065637+DorianZheng@users.noreply.github.com>
Co-authored-by: Copilot Autofix powered by AI <223894421+github-code-quality[bot]@users.noreply.github.com>

Author: DorianZheng

apps/infra-local/CONNECTIONS.md

@@ -1,99 +1,34 @@
-# BoxLite-based infra-local orchestrator
-#
-# All targets are thin wrappers around `python -m boxlite_local`. The Python
-# CLI is the source of truth — see `python -m boxlite_local --help` for the
-# full surface area. This Makefile exists for muscle memory + tab completion.
+# infra-local — thin aliases over `python -m compose <cmd>`.
+# The Python CLI is the source of truth: `python -m compose --help`.
+# This Makefile exists for muscle memory + tab completion.
 
 PY ?= python
 
-# All recipes must agree with scripts/_stack-common.sh (and the python
-# orchestrator's own default) on where the L1 SDK home lives, so the
-# `boxlite` CLI here (stack-rebuild-l1-box) sees the same boxes.
-REPO_ROOT := $(abspath $(CURDIR)/../..)
-export BOXLITE_HOME ?= $(REPO_ROOT)/.apps-local/boxlite
+.PHONY: help install up down status logs restart reset nuke
 
-# Prefer a boxlite CLI built from THIS checkout over a stale global install
-# (an older PATH `boxlite` can fail to read this version's box-config rows).
-# Matches the resolver in scripts/_stack-common.sh.
-BOXLITE_CLI := $(firstword $(wildcard $(REPO_ROOT)/target/release/boxlite $(REPO_ROOT)/target/debug/boxlite) boxlite)
+help:      ## show this help
+	@awk -F':.*## ' '/^[a-zA-Z][a-zA-Z0-9_-]*:.*## / { printf "  %-9s %s\n", $$1, $$2 }' $(MAKEFILE_LIST)
 
-.PHONY: help up down ps doctor logs wipe itest test install migrate e2e itest-all \
-        stack-build stack-up stack-down stack-restart stack-status stack-logs stack-reset stack-reset-hard stack-nuke \
-        seed-init-data stack-rebuild-l1-box
+install:   ## install the package (editable; pulls the boxlite SDK)
+	$(PY) -m pip install -e .
 
-help:                  ## show this help
-	@awk -F':.*## ' '/^[a-zA-Z][a-zA-Z0-9_-]*:.*## / { printf "  %-16s %s\n", $$1, $$2 }' $(MAKEFILE_LIST)
+up:        ## ensure L1 boxes + start L2 (COMPONENTS="api runner")
+	$(PY) -m compose up $(COMPONENTS)
 
-install:               ## install the package + test extras
-	$(PY) -m pip install -e ".[test]"
+down:      ## stop L2 procs (ARGS=--all also stops L1 boxes)
+	$(PY) -m compose down $(ARGS) $(COMPONENTS)
 
-up:                    ## bring all services up (runs doctor preflight first)
-	$(PY) -m boxlite_local up
+status:    ## one-screen L1 + L2 health
+	$(PY) -m compose status
 
-down:                  ## stop L2 native procs + remove L1 services
-	-./scripts/stack-down.sh
-	$(PY) -m boxlite_local down
+logs:      ## tail a component log (COMPONENT=api|all)
+	$(PY) -m compose logs $(COMPONENT)
 
-wipe:                  ## stop L2 + remove L1 + wipe data dir
-	-./scripts/stack-down.sh
-	$(PY) -m boxlite_local down --wipe
+restart:   ## restart L2 proc(s) / recreate L1 box(es) (COMPONENTS="runner dex")
+	$(PY) -m compose restart $(COMPONENTS)
 
-ps:                    ## list running boxlite-local-* boxes
-	$(PY) -m boxlite_local ps
+reset:     ## wipe L2 runtime state (ARGS=--hard also rebuilds the schema)
+	$(PY) -m compose reset $(ARGS)
 
-doctor:                ## run preflight checks (SDK + runtime + port conflicts)
-	$(PY) -m boxlite_local doctor
-
-test:                  ## run unit tests (no BoxLite required)
-	$(PY) -m pytest tests/unit -q
-
-itest:                 ## run integration smoke test (requires BoxLite runtime, ~30s)
-	BOXLITE_INTEGRATION=1 $(PY) -m pytest tests/integration/test_multi_service.py -v -s
-
-e2e:                   ## run comprehensive E2E suite (~60s, ~90s with smoke)
-	BOXLITE_INTEGRATION=1 $(PY) -m pytest tests/integration/test_e2e_full.py -v -s
-
-itest-all:             ## run BOTH integration suites (~90s)
-	BOXLITE_INTEGRATION=1 $(PY) -m pytest tests/integration -v
-
-migrate:               ## build local pg schema by running all TypeORM migrations from scratch
-	cd ../api && set -a && . ../.env && set +a && \
-	  npx ts-node -P ./tsconfig.json -r tsconfig-paths/register \
-	    ../node_modules/typeorm/cli.js migration:run -d ./src/migrations/data-source.ts
-
-seed-init-data:        ## ensure dashboard-required base data (admin org, default region, wait runner)
-	./scripts/seed-init-data.sh
-
-# ── L2 stack wrappers (L1 + native API/Runner/Proxy/Dashboard) ───────────────
-
-stack-build:           ## build native runner + proxy binaries
-	./scripts/stack-build.sh
-
-stack-up:              ## ensure L1 up + start all L2 native processes (idempotent)
-	./scripts/stack-up.sh $(COMPONENTS)
-
-stack-down:            ## stop all L2 native processes (L1 stays up; add ARGS=--all to stop L1 too)
-	./scripts/stack-down.sh $(ARGS) $(COMPONENTS)
-
-stack-restart:         ## restart one or more L2 components (runner = also rebuild)
-	./scripts/stack-restart.sh $(COMPONENTS)
-
-stack-status:          ## one-screen health check across L1 + L2
-	./scripts/stack-status.sh
-
-stack-logs:            ## tail logs (COMPONENT=api|runner|proxy|dashboard|all)
-	./scripts/stack-logs.sh $(COMPONENT)
-
-stack-reset:           ## wipe L2 runtime state (PG user data + runner home; L1 + schema preserved)
-	./scripts/stack-reset.sh
-
-stack-reset-hard:      ## like stack-reset, but also re-applies prod schema baseline
-	./scripts/stack-reset.sh --hard
-
-stack-nuke:            ## ABSOLUTE NUKE: L1 boxes destroyed + data wiped + logs cleared
-	./scripts/stack-reset.sh --nuke
-
-stack-rebuild-l1-box:  ## destroy + recreate one L1 box (BOX=dex|registry|...) — for stuck stateful services
-	@if [ -z "$(BOX)" ]; then echo "usage: make stack-rebuild-l1-box BOX=<name>"; echo "e.g. BOX=dex   when login returns stale tokens"; echo "e.g. BOX=registry  when image pulls hang"; exit 1; fi
-	$(BOXLITE_CLI) rm boxlite-local-$(BOX) --force
-	python -m boxlite_local up $(BOX)
+nuke:      ## tear down EVERYTHING (L1 boxes + data + logs)
+	$(PY) -m compose nuke