[ET-VK][int4] patch 4-bit source transformation quantizer to support linear modules with biases

Pull Request resolved: https://github.com/pytorch/executorch/pull/8224

While LLaMa does not have biases, there are some models which will have biases in their linear modules.

Add support in the source transform quantizer for biases.
ghstack-source-id: 264952608
@exported-using-ghexport

Differential Revision: [D69072087](https://our.internmc.facebook.com/intern/diff/D69072087/)

Co-authored-by: Nathanael See <nathanaelsee@fb.com>