Bring tile-consumer-and-fuse-producers pass from TPP

[dchigarev]

Closes #194

Copies tile-consumer-and-fuse-producers pass required for linalg->xegpu->gpu_exe pipeline from tpp-mlir repo.

The pass tiles linalg ops. An example of how this pass works.

[ZhennanQin]

I suggest reusing the CPU implementation from #87, which would do a similar transformation. Would you please discuss with @Yun-Fly which kind of transformation do you require and he can enable that in that PR.

[Yun-Fly]

I suggest reusing the CPU implementation from #87, which would do a similar transformation. Would you please discuss with @Yun-Fly which kind of transformation do you require and he can enable that in that PR.

+1, Agree with @ZhennanQin.

In CPU pipeline, we have enabled a new fusion pass, which has more fusion capability than tile-consumer-and-fuse-producers. You can refer to #87 for more information.

If you have any questions about this, please let me know.

BTW, the example you illustrated looks more like a tiling process instead of fusion, because there are no other ops except for matmul.