Add static switch utility and utility functions for tensor operations

- Introduced `static_switch.h` with BOOL_SWITCH, EVENK_SWITCH, SOFTCAP_SWITCH, FP16_SWITCH, and HEADDIM_SWITCH macros for compile-time conditional execution.
- Added `utils.h` containing various utility functions for tensor operations, including relu, max, sum, and GEMM implementations.
- Implemented specialized relu functions for half and bfloat16 types using inline PTX assembly for performance optimization.
- Enhanced tensor layout conversion functions to support different configurations for GEMM operations.
- Included support for asynchronous copy operations and softmax calculations within the FLASH_NAMESPACE.

Author: LoserCheems