[CodeGen][CUDA] Move fast math intrinsic lowering option to PassContext

[tlopex]

This updates CUDA fast math intrinsic lowering to use a PassContext option instead of a CUDA Target attribute.

The new option is:

with tvm.transform.PassContext(config={"tirx.enable_fast_math": True}):
    ...

When unset or false, CUDA math intrinsics continue to lower to the precise CUDA math functions such as expf. When true, tirx.LowerIntrin prioritizes the cuda.fastmath.* lowering rules, producing fast math intrinsics such as __expf.

[gemini-code-assist]

Code Review

This pull request refactors the enable_fast_math setting by moving it from a CUDA target attribute to a global pass configuration option (tirx.enable_fast_math). The changes include removing the attribute from target detection, registration, and the CUDA target kind definition, while updating the LowerIntrin pass and associated tests to utilize the PassContext for this configuration. I have no feedback to provide as there were no review comments to evaluate.