[CI] Add TIRX tests to CI

[tlopex]

This PR adds tests/python/tirx to the general Python unittest CI.

Previously, task_python_unittest.sh only ran:

• tests/python/tirx-analysis
• tests/python/tirx-base
• tests/python/tirx-transform

but did not run the TIRX DSL/operator/codegen tests under tests/python/tirx.

This PR adds tirx to TEST_FILES and marks CUDA-only TIRX tests with requires_cuda, so CPU CI can run the full directory while skipping CUDA-only cases. It also adds compute capability guards for Hopper/Blackwell/TMA/TCGEN-specific tests to avoid running them on unsupported GPU architectures.

[gemini-code-assist]

Code Review

This pull request enables CUDA testing for the tirx component by adding pytest marks and compute version requirements across multiple test files, and including tirx in the unit test execution script. The feedback provided identifies several opportunities to improve the precision of these guards, specifically by applying module-level compute version requirements for architecture-specific features like Hopper (SM 90) and Blackwell (SM 100), and correcting the version argument format for SM 7.5.

[tlopex]

/gemini review

[gemini-code-assist]

Code Review

This pull request refactors CUDA compute version requirements in the TVM testing utility by introducing requires_cuda_compute_version_marks and updating the requires_cuda_compute_version decorator. These changes are applied across numerous TIR-X test files to enforce CUDA requirements at the module level using pytestmark. Additionally, the tirx test suite is now included in the unit test execution script. The review feedback correctly identifies several instances where function-level decorators have become redundant due to the new module-level marks and should be removed to improve code maintainability.