Assortment of optimizations

[ChrisPenner]

Overview

Adds a couple optimizations and some debugging stuff.

• Adds deepTime which times not only the monadic action, but also deeply forces the returned argument. This gives a much more accurate depiction of the amount of time a given thing takes.
• Removes unsafeTime, I tested it, and as we presumed, it's just broken.
• Adds enough NFData instances for me to use deepTime on a Names object, since that's where most of my profiling work has been lately 🙃 ; we can add more as necessary.
• Fixes some Debug weirdness where pTrace'ing strings would just erase part of them.
• Adds deepEvaluate to the Debug module, it can help with determining when something is slow.
• Speeds up TDNR lookup from O(freeVars * names) to ~freeVars * log(names)
• Replaces a linear Names.filter in slurping with a set difference operation

Implementation notes

The only one that has an interesting implementation is findMatchingTermSuffixes; so I'd recommend checking that out.

Test coverage

Benchmarked against 0.5.24, I cloned nimbus, then edit.namespace, here's the time it took to parse the scratchfile:

• ucm (0.5.23) :
  • First run: 29.1 s (cpu), 22.3 s (system)
  • Second run: 25.1 s (cpu), 19.0 s (system)
• cp/project-root:
  • First run: 21.9 s (cpu), 14.8 s (system)
  • 12.3 s (cpu), 8.73 s (system)
• cp/optimizations-2 (optimized):
  • First run: 16.0 s (cpu), 15.1 s (system)
  • Second run: 10.3 s (cpu), 9.90 s (system)

So not a crazy difference from project-root, but still saves us a second or so.

[mergify]

⚠️ The sha of the head commit of this PR conflicts with #5041. Mergify cannot evaluate rules on this PR. ⚠️

[ChrisPenner]

Calling pTrace on something that's already a string, tries to parse it like Show output and ends up hiding bits and pieces of it randomly :'(

[aryairani]

@ChrisPenner Any idea why CI / run transcripts (windows-2019) (pull_request) failed on reflog transcript? Shouldn't that include the code that hides it?

[ChrisPenner]

Oh strange; I wonder if it inserted them so close that they have the same timestamp and sqlite is sorting them out of order 🤔

BTW not even sure if this PR is worth merging, conceptually this should be a speed up, but the real-world numbers aren't super significant.