TPP Runner Wrapper pass

[adam-smnk]

Refactor MLIR Bench infrastructure into a pass.

The core logic is already covered by tpp-run tests.

[adam-smnk]

For TPP+IMEX testing, I settled on tpp-run but running tpp-opt with desired part of the pipeline is overall cleaner.
Also, creating runner wrapper is pretty useful and best done before any other transformations. So, this refactor exposes kernel wrapper creation to tpp-opt too.

[rengolin]

The pass name is -mlir-bench but the directory is Runner and the library built is TPPRunner. Is this a generic MLIR benchmark pass or a TPP specific thing? If the former, the naming needs to be more generic. If the latter, why change?

Do you aim to upstream this?

[adam-smnk]

Do you aim to upstream this?

No, I don't think it's generic enough for that. It stays internal for now.

If the latter, why change?

Mostly for future integration with external tools. Change is motivated by the current use case of TPP -> IMEX pipeline:
tpp-run INPUT FLAGS | sed '/Error:.*/d' | IMEX.
The error is produced because we try to run an incomplete lowering but I want to use tpp-run for simplicity.

This separate runner wrapper pass allows to retain our handy wrappers and avoid silly errors when terminating lowering pipeline early. With this PR, the following is possible:
tpp-opt -mlir-bench -default-pipeline | EXTERNAL CONSUMER

It might or might not be useful in the future. But the worst case, this change simplifies our runner a bit by moving the wrapper into a separate logic (pass).

Perhaps, we can give it a nicer name.

[rengolin]

I see. It's fine as is. If we want to upstream later, we can clean up.

[adam-smnk]

NFC rebase on main + renamed to tpp-runner-wrapper