feat: one-command local dev stack (make localnet) with seed, monitoring, and a fresh-init migration fix

[addnad]

Summary

One-command local development environment, per the scope confirmed on the issue (#210 comment): make localnet boots the API + TimescaleDB Postgres (migrations auto-applied) + OTel collector + Prometheus + Grafana (pre-provisioned with the repo's existing dashboards), then seeds validators and mock telemetry. make localnet-clean removes all state.

make localnet         # build + boot + seed
make localnet-logs    # tail services
make localnet-clean   # tear down, remove volumes

What's included

• deploy/localnet/docker-compose.yml — postgres (timescale/timescaledb-ha:pg16, with pg_cron preloaded so migration 002 can install it), otel-collector (OTLP gRPC :4317), api (built from the repo Dockerfile), prometheus (scrapes the API's /metrics), grafana (:3001, anonymous viewer, Prometheus datasource + the dashboards from deploy/observability/ and deploy/monitoring/ file-provisioned).
• scripts/seed-localnet.ts — idempotent seed: a bond pool + 8 validators, each with a stake (validator_stakes), reputation (reputations), and pending rewards (reward_pending_amounts) — the "pre-funded accounts" equivalent in this schema.
• scripts/mock-telemetry.ts — emits a window of uptime_heartbeat rows + reward_tx per validator — the "mock attestation" equivalent.
• Makefile — localnet, localnet-seed, localnet-mock, localnet-logs, localnet-clean. Host port for Postgres is overridable (LOCALNET_PG_PORT) if 5432 is taken.
• deploy/localnet/README.md — usage + the scope mapping.

Pre-existing migration bug found & fixed (blocks every fresh DB init)

Live-testing the stack surfaced that 011_distributed_jobs.sql fails on any fresh database: its partial index used NOW() in the predicate —

WHERE status IN ('pending','running') AND (locked_until IS NULL OR locked_until <= NOW());

Postgres rejects this (functions in index predicate must be marked IMMUTABLE) because index membership is fixed at write time while NOW() changes continuously — so the migration aborts under initdb and under the app's own migration_manager (same SQL). The fix keeps the static status filter (which excludes the bulk of the table) and leaves the time-dependent locked_until check at query time, where lease queries already apply it and can still use the index.

Verification (all run live)

• Fresh container init applies all 14 migrations with 0 errors; timescaledb + pg_cron extensions active; the fixed idx_distributed_jobs_ready exists.
• seed-localnet → 8 validators seeded (verified validator_stakes count = 8).
• mock-telemetry → 248 heartbeat rows + 24 reward_tx confirmed in the DB.
• docker compose build api (repo Dockerfile: npm ci + tsc) → builds clean.
• docker compose config validates.

Closes #210