[CI][TIRx] Enable compile coverage and fix exposed regressions

[tlopex]

This PR removes the suite-wide TIRx hardware gate, splits mixed CUDA tests into compile and GPU-run variants so regular CI always compiles kernels against explicit target architectures, and skips execution only after compilation when matching hardware is unavailable. It also fixes the parser, type handling, CUDA intrinsic, TMA, FP4/FP8 NVRTC, and shared-memory regressions exposed by the newly enabled compile coverage.

[gemini-code-assist]

Code Review

This pull request removes the suite-level hardware gate for TIRX tests, updates target handling in CUDA codegen tests, and adds @pytest.mark.gpu decorators to several test cases. Feedback focuses on correcting an inaccurate return type annotation in _get_source and removing the @pytest.mark.gpu decorators from compile-only tests. Since these compile-only tests do not execute on physical hardware, removing the decorators and specifying concrete target architectures (like sm_90a or sm_100a) inside the tests will allow them to run in CPU-only CI environments.

Important

The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.

[tlopex]

@tvm-bot rerun