Replace Linear lowering of using Matmul with Conv2d (#1336)

Summary:
- The existing Vela compiler doesn't support Matmul and TOSA.Fully_Connected will be deprecated
- Also add support for linear layers with rank>2 + tests

Pull Request resolved: https://github.com/pytorch/executorch/pull/1336

Reviewed By: cccclai

Differential Revision: D51922063

Pulled By: digantdesai

fbshipit-source-id: a8ad11f170911c1543fe88fd2ef1a1356ac859c3