ml/: no quality-gate script runs Python unit tests (scripts/validate-ml-tests missing) #245

Context

Discovered while implementing ADR-340 (PassFilter/VariationGenerator randomization primitives, ml/pipeline/core/randomization.py).

scripts/validate.sh runs three gates today:

validate-build.sh      # dotnet build
validate-ml-build.sh   # mypy --strict over ml/pipeline and ml/test
validate-tests.sh      # dotnet test

There is no scripts/validate-ml-tests.sh counterpart that actually executes the Python unit tests landing in ml/test/. pytest is available in the dev environment and is pinned in ml/requirements.txt (added on the ADR-339 branch), and ml/test/pipeline/core/test_randomization.py (ADR-340) plus ml/test/pipeline/core/test_manifest.py (ADR-339) both already pass when run manually (cd ml && python3 -m pytest test/), but nothing in the project's own quality-gate pipeline runs them.

This is a gap against CLAUDE.md's Quality Gates section, which states a change isn't complete until scripts/validate-tests (all unit and E2E tests) passes — that gate currently only covers dotnet test, with no way to enforce the ADR-191 epic's "unit tests written test-first" exit criterion (required on every task from ADR-339 onward) via the standard pipeline.

Suggested fix

• Add scripts/validate-ml-tests.sh/.cmd (mirroring the validate-ml-build naming/pattern) that runs pytest from ml/.
• Wire it into scripts/validate.sh/.cmd alongside validate-ml-build.
• Update CLAUDE.md's Quality Gates table to list it.

Scope note

Per ml/_spec_OopPipeline.md's task breakdown (ADR-338 through ADR-358), no task currently owns this wiring — it fell out of scope for ADR-340 (whose task brief explicitly scoped file changes to ml/pipeline/core/randomization.py and its test file only) and wasn't added by the sibling ADR-339 task either. Filing this so it's tracked and picked up as its own small task rather than silently expanding an in-flight task's scope.