`addmm` and `baddbmm` (#3401)

Summary:
Pull Request resolved: https://github.com/pytorch/executorch/pull/3401

We modify the `matmul` shader to support `addmm`.  `baddbmm` is therefore supported since it is decomposed as
```
aten.add.Tensor
aten.mul.Tensor
aten.mul.Tensor
aten.bmm.default
```

Reviewed By: SS-JIA

Differential Revision: D56678410

fbshipit-source-id: 43bdc051eca9c8d1bbbfaf63c4ba81cc8cb3cd2b