[Dev] Enhance Operator Cache to support multi-thread environments

[LeiWang1999]

Fix ref to #204 #186

This pull request introduces several changes to improve thread safety, enhance scheduler functionality, and refactor code for better readability and maintainability. The most important changes include adding a lock to the OperatorCache class, modifying the ThreadPoolExecutor to use a variable number of workers, and introducing a new fine-grained matrix multiplication scheduler.

Thread Safety Enhancements:

• bitblas/cache/operator.py: Added a cache_locker using threading.RLock to synchronize access to the cache in methods like add, get, clear, and save_into_database. [1] [2]

Scheduler Improvements:

• bitblas/base/utils.py: Modified ThreadPoolExecutor to use a variable number of workers (max_workers) instead of a fixed number (4).
• bitblas/ops/base_scheduler.py: Added a method get_hardware_aware_configs to raise a NotImplementedError for hardware-aware tuning.
• bitblas/ops/general_matmul/tilelang/dense/matmul_simt.py: Introduced a new MatmulFineGrainSIMTScheduler class for fine-grained matrix multiplication scheduling.

Code Refactoring:

• bitblas/ops/general_matmul/tilelang/dense/matmul_tensorcore.py: Renamed from matmul.py and added imports and methods for hardware-aware configurations. [1] [2] [3]
• bitblas/ops/operator.py: Refactored multiple methods for better readability, including apply_fast_tuning, hardware_aware_finetune, and _build_default_module. [1] [2]

Additional Changes:

• bitblas/ops/general_matmul/tilelang/dense/__init__.py: Updated imports to include matmul_simt and matmul_tensorcore.
• bitblas/ops/operator.py: Added import for tl_apply_and_build from bitblas.tl.tuner.