Audit tests for @pytest.mark.gpu coverage

[Abdelsalam-Abbas]

Summary

Review the test suite and add @pytest.mark.gpu to tests that should run on GPU runners but are currently missing the marker.

Context

PR #156 introduced the @pytest.mark.gpu marker and applied it to the obvious candidates. @marcuscollins suggested a follow-up pass to catch any tests we missed.

Things to consider:

• Tests using try_gpu fixture don't require GPU (they fall back to CPU), but some may benefit from running on GPU
• Any new tests added for GPU-dependent features should get the marker
• gpu-2 and gpu-4 runners are available if heavier tests are needed in the future

Acceptance criteria

• All tests that require or strongly benefit from GPU have @pytest.mark.gpu
• pixi run cpu-tests passes without GPU hardware
• pixi run gpu-tests captures all GPU-dependent tests