feat: Exhaustive greedy rewrite strategy

[aborgna-q]

This is a straight combination of the greedy and exhaustive strategies. Instead of applying each strategy separately, apply each one once but also include all the following strategies that do not modify the same nodes.

This is a draft, since for some reason combining multiple strategies at once is invalidating the hugr.
They may not be correctly reporting their modified nodes ?

[aborgna-q]

This requires CQCL/hugr#600 to check whether the rewrites are non-overlapping.

We get really nice results.

On my laptop, 1 thread, barenco_tof_10, 60s timeout, queue size 500:

Base cost: 192 CX gates, 326 gates.

• Using HEAD:

  • Min cost: (135, 269)
  • 281 circuits processed / 28k seen
  • Peak memory usage: 508MB

• With this branch:

  • Min cost: (120, 260)
  • 340 circuits processed / 24k seen
  • Peak memory usage: 427 MB

It's specially noticeable on short timeouts; we beat HEAD with cost (132, 268) in 5 seconds.

[lmondada]

Excellent!