Pccm

Fixed
- fix a bug that bool is captured as int64_t in inliner.
- fix a bug of additional vars in inliner.

Added
- add compiled conversion to greatly reduce launch overhead of inline function. currently only inline cuda/metal kernels in cumm support this.

Fixed
- fix a bug in simple type analysis
Added
- add metal constant filter in type analysis
- add numpy scalar type support in type analysis

Added
- add attributes and `code_before_func_def` support for apple metal

Added
- add support for param class reload by check id in `sys.module`

Fixed
* fix wrong version.txt