Model-quantizer

Changed
- Significantly reduced core dependencies to minimize installation issues
- Moved non-essential dependencies to optional extras
- Created new extras: 'viz' for visualization and 'data' for dataset handling
- Changed gptqmodel dependency to use versions below 2.1.0 to avoid numpy>=2.2.2 requirement
- Improved requirements.txt with clearer organization and comments
- Removed torch as a direct dependency to allow more flexible installation

Fixed
- Pinned dependency versions to match working Python 3.11 environment
- Updated torch to version 2.5.1
- Updated bitsandbytes to version 0.42.0
- Updated gptqmodel to version 2.1.0
- Added torchvision and torchaudio as explicit dependencies
- Added "all" extra in setup.py to install all dependencies at once

Fixed
- Added explicit gptqmodel dependency for GPTQ quantization
- Fixed issue with transformers reporting gptqmodel as available when it's not installed
- Added proper requirements.txt file with all dependencies
- Improved dependency management in setup.py

Fixed
- Completely redesigned Python 3.12 compatibility for GPTQ quantization
- Added multiple patching strategies for transformers 4.49.0 compatibility
- Implemented recursive function scanning to find and patch CUDA checks
- Added method-level exception handling to bypass GPU requirements on CPU

Fixed
- Improved Python 3.12 compatibility for GPTQ quantization
- Fixed patch targeting for CUDA availability check in transformers
- Added multiple fallback methods to ensure CPU compatibility

Fixed
- Added Python 3.12 compatibility for GPTQ quantization
- Fixed issue with Optimum's CUDA check in Python 3.12
- Applied monkey patch to bypass CUDA requirement in newer Python versions