Add bias support to QQLinear

[mdepree]

Proposed changes

Currently, QQLinear does not support bias terms, which prevents quantization of models that use biased linear layers. This limitation is noted in the existing docstring: "Note: This layer does not support a bias term yet." This PR adds support for an optional bias term to the QQLinear layer, bringing it in line with Linear and QuantizedLinear.

Checklist

Put an x in the boxes that apply.

• I have read the CONTRIBUTING document
• I have run pre-commit run --all-files to format my code / installed pre-commit prior to committing changes
• I have added tests that prove my fix is effective or that my feature works
• I have updated the necessary documentation (if needed)

[nastya236]

Thanks for adding this. Unfortunately, the PR is incomplete. Adding a bias requires changes to how layouts are initialized for QQ matmul in mlx/backend/cuda/quantized/cublas_qqmm.h. To support bias, the implementation needs to set epilogue. This should be implemented similarly to how it is handled in regular GEMM and Matmul::eval_gpu().
Happy to iterate on this with you.