GitLab

* [Enhancement] Add NVCC support for PTX and SASS generation in TileLang

* Introduced functions to compile CUDA C++ source to PTX and SASS formats, enhancing the ability to generate intermediate representations for CUDA kernels.
* Added default compile options for NVCC, including paths for TileLang templates, CUTLASS, and CUDA includes.
* Implemented methods to export and display generated PTX and SASS code, improving usability for developers working with CUDA targets.
* Updated JITKernel class to integrate new NVCC functionalities for PTX and SASS handling, ensuring compatibility with existing workflows.

* [Fix] Improve error handling in get_sass_from_source function

* Added contextlib to suppress exceptions when removing temporary files, enhancing robustness.
* Fixed formatting of error message for clarity when CUDA tools are not found, ensuring better user feedback.

* [Enhancement] Preserve user flags in NVCC compile options

* Updated the default_compile_options function to preserve user-specified compile flags, including repeated tokens, by utilizing shlex for proper tokenization.
* This enhancement improves the flexibility and accuracy of NVCC compile options, ensuring that all user inputs are correctly handled.