Use memoized variants for mapping expressions

[kaushikcfd]

See inducer/pymbolic#90.

Draft because:

• Waiting for Add memoized variants of mappers pymbolic#90 to be complete.
• Pass CI
• Contains commits from Avoid mutable arguments in CSEToAssignmentMapper #610

[inducer]

Something like #611 (currently set to automerge if it passes) may address the failure in the Firedrake CI.

[kaushikcfd]

Something like #611 (currently set to automerge if it passes) may address the failure in the Firedrake CI.

Thanks, pulled from main. Already noticed an oddity: Pytential CI in the last attempt passed, but the CI times went up by 50%.

[inducer]

Pytential CI in the last attempt passed, but the CI times went up by 50%.

Just this PR, or one of the other ones?

[inducer]

• fix map_comparison and map_substitution in IdentityMapper pymbolic#97 has a 43 minute pytential time.
• https://github.com/inducer/loopy/runs/6284535897?check_suite_focus=true from Firedrake CI: update pymbolic #611 has 36 min.

[kaushikcfd]

I think the variance between runs is pretty high, an older run: inducer/pymbolic#91 (a week ago) has 49 mins of potential CI time. So, not surprising quite surprising in that case. Looks mostly like a false alarm. (sorry)

[inducer]

This looks good to me. Thanks for working on it!