Width Packing Mat1 input for Quantized Linear (#6149)

Summary:
Pull Request resolved: https://github.com/pytorch/executorch/pull/6149

Width packing mat1 input for Quantized Linear as ASR model provides channel-packed matrix while operator does not support channel-packed yet.

Reviewed By: nathanaelsee, jorgep31415

Differential Revision: D64065606

fbshipit-source-id: 2a7d43d432deef7245d1d45f5c760b0f42627551