bench: add back benchmark and profiling infrastructure

[EdmundGoodman]

Summary

This PR moves the existing benchmarks and profiling infrastructure from https://github.com/xdslproject/xdsl-bench into the main xDSL repo so they can be used more easily by developers. In addition to this, it restructures and extends them to provide more microbenchmarks, component-level benchmarks, and end-to-end benchmarks.

A full description of the motivate and structure is available in benchmarks/README.md.

Future work

1. Modify https://github.com/xdslproject/xdsl-bench to run against new benchmark repo (blocked by merging this PR)
2. Add further benchmarks, with main areas for development being (to be done as needed within the provided framework)
   • Evaluate additional components, such as the printer and verifier
   • Evaluate larger "hero" end-to-end workloads, such as CIRCT, ASL, or quantum
   • Make synthetic data driving microbenchmarks more representative
3. Consider reducing the amount of MLIR duplicated from the test directory, by leveraging the mlir-opt installation in CI to generate it on the fly

[superlopuh]

Is there a way to avoid duplicating the files? I was expecting this to be more minimal

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 91.26%. Comparing base (dab099d) to head (d577040).
Report is 3 commits behind head on main.

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #3889   +/-   ##
=======================================
  Coverage   91.26%   91.26%           
=======================================
  Files         466      466           
  Lines       57949    57949           
  Branches     5575     5575           
=======================================
  Hits        52886    52886           
  Misses       3635     3635           
  Partials     1428     1428           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[EdmundGoodman]

Is there a way to avoid duplicating the files? I was expecting this to be more minimal

If I remove the benchmarks which parse/lex over all test MLIR files instead of individual ones that should cut down the size a lot?

[EdmundGoodman]

OK, this looks a bit more sensible in the number of files changed and shouldn't impact the expressiveness of the tests too much.

[compor]

this .. feels a bit fragile, especially repeated at every make target.

Can't we use the --directory option along with exposing this as a make var along the whole Makefile?