[Dev] Refactor Backend Dispatch and Kernel Wrap Related Design

[LeiWang1999]

This pull request mainly includes changes to improve the architecture of the project, modify the CI workflow, and update the third-party notices. The most significant changes are the addition of the LibraryGenerator class, the modification of the BaseWrapper class, and the changes in the import statements to reflect the new project structure.

Improvements to project architecture:

• bitblas/base/roller/__init__.py, bitblas/base/roller/policy/default.py, bitblas/base/roller/policy/tensorcore.py, bitblas/base/utils.py: Modified the import statements to reflect the new project structure. [1] [2] [3] [4]
• bitblas/builder/lib_generator/__init__.py: Added the LibraryGenerator class, which compiles and loads libraries, and handles the source and library paths.
• bitblas/builder/wrapper/__init__.py, bitblas/builder/wrapper/base.py: Modified the BaseWrapper class to include the wrap method. [1] [2]
• bitblas/cache/operator.py: Updated the method signatures and the method calls to use srcpath and libpath instead of src_name and lib_name. [1] [2] [3]
• bitblas/generator.py: Removed the BitBLASGenerator class.
• bitblas/ops/general_matmul/__init__.py: Renamed from bitblas/ops/general_matmul.py and updated the import statements to reflect the new project structure. Also modified the MatmulConfig class to inherit from OperatorConfig. [1] [2] [3]

CI workflow modification:

• .github/workflows/ci.yml: Modified the workflow to only run on pull requests instead of both push and pull requests.

Third-party notices update:

• THIRDPARTYNOTICES.txt: Added the notice for IST-DASLab/marlin/.

Ref to dicussion #81