[POC][WIP] Merge prologue and epilogue into main loop

[newling]

Inserting this pass after tiling is finished, but before vectorization (and before bufferization). This fails only in iree-amdaie-distribute-cores-and-objectfifos I don't think it'd be hard to get past that. I agree that making this robust to pattern matching is hard though. That said, if it were a pass in the tiling pipeline, it wouldn't need to be super robust, because the tiling pipeline isn't general purpose (we control what goes in and can annotate things).

I'm not suggesting this is a better approach than moving vectorization after bufferization and doing function outlining, I think that's good. Just throwing this option out there for future consideration

@Abhishek-Varma @jtuyls @MaheshRavishankar