[QDP] centralized Python tests and default to single-process pytest to avoid CUDA OOM

[viiccwen]

Related Issues

closes #1146

Changes

• Bug fix
• New feature
• Refactoring
• Documentation
• Test
• CI/CD pipeline
• Other

Why

This PR moves QDP Python tests into the top-level testing/qdp_python/ tree, switches the default Python test run to single-process (no -n auto), documents optional use of -n auto / -n N for faster runs, and fixes ty diagnostics for the affected tests.

How

Test layout

• Moved qdp/qdp-python/tests/test_dlpack_validation.py → testing/qdp_python/test_dlpack_validation.py
• Moved qdp/qdp-python/tests/test_quantum_data_loader.py → testing/qdp_python/test_quantum_data_loader.py

All Python tests are now under testing/ and discovered by root pytest with testpaths = ["testing"].

Makefile

• test_python now runs:
  uv run pytest --cov --cov-report=term-missing --cov-report=html:htmlcov
  (single process; coverage sources from [tool.coverage.run] in pyproject when set).

Documentation

• docs/qdp/testing.md
  • Added "Python tests" section: how to run (make test_python or uv run pytest ...).
  • Clarified that tests run in a single process by default to avoid GPU memory pressure.
  • Documented optional speed-up with -n auto or -n N when compute/GPU allow.
• qdp/DEVELOPMENT.md
  • Updated QDP Python test path from qdp/qdp-python/tests to testing/qdp_python.

Type-checking (ty)

• pyproject.toml
  • Added qumat_qdp and qumat_qdp.* to allowed-unresolved-imports so the optional qumat_qdp.loader import does not cause unresolved-import errors.
• testing/qdp_python/test_quantum_data_loader.py
  • Introduced loader_cls fixture that skips when QuantumDataLoader is not available and returns the class (for type narrowing).
  • Replaced all QuantumDataLoader(...) usages with loader_cls(...) and removed redundant assert QuantumDataLoader is not None.
  • Dropped unnecessary # type: ignore on the import fallback.
• testing/qumat/test_parameter_binding.py
  • Assigned get_backend_config(backend_name) to a variable, added assert cfg is not None, then used cfg.copy() to satisfy ty (possibly-missing-attribute on optional dict).

Other

• uv.lock
  • Reflects current dependency resolution (e.g. qumat-qdp editable); any change shown is from lock refresh.

Checklist

• Added or updated unit tests for all changes
• Added or updated documentation for all changes

[ryankert01]

Thanks for the fast fix! Some comments

[ryankert01]

I looked into it and discovered the peak memory is around 10gb! It runs blazingly fast on a gpu server(A6000).

[rich7420]

@viiccwen take a look about conflicts

[ryankert01]

need resolve the conflicts