DQX Studio: Replace apx framework with first-party build/dev scripts

[berrybluecode]

The app previously relied on the unpublished apx package (github.com/databricks-solutions/apx) to drive its build and dev workflows. Drop the dependency and reimplement what we need ourselves:

• scripts/build_app.py: replaces apx build. Generates the FastAPI OpenAPI schema, runs orval (now configured via orval.config.ts) to regenerate ui/lib/api.ts, runs vite build, and produces the application wheel + app.yml + requirements.txt under .build/. The wheel carries a build-tag local-version segment (e.g. .b20260530t012345) baked into both the filename and the internal METADATA, so successive deploys at the same git commit always reinstall fresh code in the Databricks Apps persistent venv. This obsoletes scripts/_align_wheel_version.py (now removed), which used to repair apx's filename-only mangling after the fact.

• scripts/dev.py: replaces apx dev start. Runs uvicorn with --reload on a fixed internal port and serves the SPA via Vite on :9001, proxying /api and /docs to the backend. SIGINT is forwarded to both children and either child crashing tears the other down, so no orphan processes are left behind.

• pyproject.toml: drop the apx dependency-group entry and the [tool.uv.sources].apx git source. Rename the metadata table from [tool.apx.metadata] to [tool.dqx_app.metadata] (same schema, new namespace) and document its consumers (build_app.py, vite.config.ts). Add basedpyright to the dev group.

• databricks.yml: rewrite artifacts.default.build to call scripts/build_app.py instead of apx build. Add explicit Mode A (bundle-managed) vs Mode B (external/reuse) documentation on the sql_warehouse_id variable; default to empty so a target that forgets to set it fails fast.

• post_deploy_grants.sh: extend the warehouse grant path so external warehouses (Mode B) get CAN_USE granted to the app SP, job SP, and the users group via the warehouses permissions API — the bundle doesn't own the warehouse in that mode, so the bundle's permissions: block can't do it.

• Documentation refresh across DEPLOYMENT.md, DEVELOPMENT.md, README.md, AGENTS.md, the UI/backend CLAUDE.md files, and docs/dqx/docs/{installation,dev/contributing}.mdx to reflect the new build/dev entry points and warehouse modes.

No runtime behaviour change in the app itself: app/src/...backend/app.py is touched only to pick up the new metadata namespace.

Changes

Linked issues

Resolves #..

Tests

• manually tested
• added unit tests
• added integration tests
• added end-to-end tests
• added performance tests

Documentation and Demos

• added/updated demos
• added/updated docs
• added/updated agent skills

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 92.14%. Comparing base (45bd3b1) to head (fea954e).
⚠️ Report is 2 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1223      +/-   ##
==========================================
+ Coverage   92.13%   92.14%   +0.01%     
==========================================
  Files         101      101              
  Lines        9603     9603              
==========================================
+ Hits         8848     8849       +1     
+ Misses        755      754       -1     

Flag	Coverage Δ
anomaly	53.51% <ø> (ø)
anomaly-serverless	53.52% <ø> (ø)
integration	50.05% <ø> (+1.24%)	⬆️
integration-serverless	50.70% <ø> (-0.08%)	⬇️
unit	55.92% <ø> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[github-actions]

✅ 748/748 passed, 41 skipped, 4h27m56s total

_{Running from acceptance #4805}

[github-actions]

✅ 182/182 passed, 1 flaky, 5h11m16s total

Flaky tests:

• 🤪 test_anomaly_check_with_custom_info_column_name (39.919s)

_{Running from anomaly #919}

[mwojtyczka]

Strong refactor — dropping the unpublished/un-pinned apx git dependency is a clear supply-chain win, and the scripts are unusually well-documented (the process-group signal handling in dev.py and the Mode A/B warehouse docs in databricks.yml are excellent). A few notes inline, plus two non-anchored points:

• PR claims "added unit tests" but the diff contains none. build_app.py (_read_app_metadata missing-key path, _write_metadata_module formatting, newest-wheel/stale-wheel logic) and dev.py's shutdown logic are testable. Please add a couple or correct the checklist.
• Removed apxDevProxyGuard from vite.config.ts looks correct given the new topology (Vite is now itself the public-facing proxy rather than sitting behind apx), so the origin guard no longer applies — just flagging it as a conscious drop of a (dev-only) control rather than an accident.

Nothing blocking on correctness; I'd want the grant-failure propagation and the JSON-construction hardening addressed in this deploy-privileged script before merge.

[mwojtyczka]

left some comments

[mwojtyczka]

LGTM