Rewrite as production-ready Python package (v0.1.0 on PyPI)

[trouze]

Summary

Replaces the prototype with a proper src/-layout Python package intended to ship on PyPI as dbt-dag-opt. The package name is available; first release will be v0.1.0.

• CLI: Typer-based dbt-dag-opt analyze with file-mode (--manifest/--run-results) and dbt Cloud mode (--account-id/--job-id/DBT_CLOUD_TOKEN), plus --format table|json|jsonl, --top N, --output, and --run-id for historical runs.
• Algorithm: single iterative DP over topological order replaces the per-source recursive DFS + ProcessPoolExecutor. O(V+E) across all sources. Removes the 20s-per-task timeout ceiling and the recursion-limit risk.
• Correctness fixes carried from the prototype:
  • Node weights are now attached to the target of each hop (prototype used parent-weight on outgoing edges — silently wrong for any branching DAG).
  • if node not in self.weights bug (compared node ids against a list of floats) is gone with the rewrite.
  • Output is now valid JSON (single object keyed by source) instead of the prior append-mode stream of comma-joined {key: value} fragments.
• Testing: 29 pytest cases across artifact loaders (incl. responses-mocked Cloud API), graph building, longest-path correctness on diamonds / linear chains / cycles / single-node / missing-run-result, formatters, and the CLI via typer.testing.CliRunner.
• Tooling: ruff, mypy --strict, py.typed, hatchling build backend, uv.lock committed for reproducible CI.
• CI/CD: ci.yml matrix on Python 3.10 / 3.11 / 3.12 (lint + type + test + coverage). publish.yml is tag-triggered and uses PyPI Trusted Publishing (OIDC) — no long-lived tokens in repo secrets.
• Docs: rewritten README with install, quickstart, CLI reference, sample output, and explicit out-of-scope notes for v0.1.0. CHANGELOG.md documents the upgrade.

Deferred to later versions

• v0.2: thread-aware scheduler simulation. Real wall-clock is bounded by both critical path and threads; v0.1 treats the critical-path distance as a lower bound.
• v0.3: warehouse cost model (--warehouse-size, --rate $/hr → projected $ savings for Snowflake-style billing).
• multi-run averaging (stabilize weights across N recent runs), graphviz / mermaid export.

Test plan

• uv run ruff check . — clean
• uv run mypy src — clean under --strict
• uv run pytest — 29/29 passing
• uv build + uv run --with twine twine check dist/* — both sdist and wheel pass
• uv run dbt-dag-opt --version → dbt-dag-opt 0.1.0
• uv run dbt-dag-opt analyze --manifest tests/fixtures/tiny_manifest.json --run-results tests/fixtures/tiny_run_results.json --format table — top path is raw.orders → stg_orders → int_orders → fact_orders at 35.00s, matching hand-verified expectation
• Manual before tagging v0.1.0: run against a real manifest.json + run_results.json from a past dbt project and eyeball top paths
• Before first tag push: configure PyPI Trusted Publisher (Project → Publishing → GitHub → trouze/dbt-dag-opt / publish.yml / environment pypi)

🤖 Generated with Claude Code