Refactor quantization and model loading logic; update tests for new checkpoint handling

- Removed unused variables and functions related to quantization configuration in `quantize.py`.
- Simplified the `configure_deploy` function and removed global state for weight bits.
- Updated `run.py` to streamline checkpoint loading and removed legacy code for checkpoint averaging.
- Refactored tokenizer to remove unused markers and cleaned up related constants.
- Enhanced test suite to accommodate changes in checkpoint handling and added fixtures for base archives.
- Adjusted tests to reflect the removal of mixed bit handling and updated assertions accordingly.
- Improved test coverage for model generation and embedding functionalities.