draft: Iai on CI

[mattxwang]

Trivial setup that shows how iai could be used on CI.

The Bench workflow outputs:

iai_benchmark_sdd_canonicity_c1
  Instructions:             2507685
  L1 Accesses:              4301000
  L2 Accesses:                17784
  RAM Accesses:                4114
  Estimated Cycles:         4533910

iai_benchmark_sdd_canonicity_c2
  Instructions:             2529817
  L1 Accesses:              4331311
  L2 Accesses:                17648
  RAM Accesses:                4187
  Estimated Cycles:         4566096

iai_benchmark_sdd_canonicity_c3
  Instructions:             2569170
  L1 Accesses:              4385154
  L2 Accesses:                17692
  RAM Accesses:                4264
  Estimated Cycles:         4622854

This PR also make it clear to me that we really need to cache CI, hence #66; I'm surprised that there's not a first-party cache action, but I'll poke around and see what the bigger Rust projects do.

(caching valgrind would also be nice, but this isn't the big performance hit from caching)