test: register internal and gpu pytest markers

[yaoyu-33]

Backports the pytest marker registrations from main commit e9529c3 to unblock strict-marker unit test collection on r0.5.0.

The full main commit also carries unrelated dependency, uv.lock, .main.commit, and Megatron-LM submodule changes, so this PR carries only the pyproject.toml marker hunk.

Validation:

• python3 -m pre_commit run --all-files: passed
• uv run pre-commit run --all-files: blocked on macOS ARM because nvidia-resiliency-ext==0.6.0 has no compatible wheel
• uv run python -m pytest tests/unit_tests/models/qwen_vl/modelling_qwen3_vl/test_attention.py --collect-only -q: blocked by the same dependency resolution issue

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[claude]

LGTM — clean backport. The two new marker registrations (internal, gpu) are needed because --strict-markers is enabled in addopts and both markers are already in use:

• tests/functional_tests/test_groups/training/test_distributed_save_hf_weights.py:109 — @pytest.mark.gpu
• tests/unit_tests/models/qwen_vl/modelling_qwen3_vl/test_attention.py:129 — @pytest.mark.internal

Without this registration, pytest --collect-only (and full runs) would fail on r0.5.0 with a PytestUnknownMarkWarning error.

Suggested test cases

No perf tests impacted.