10% regression in the `issue-43572-unused-uses` test case

[nikomatsakis]

The perf website shows a 10% regression in the issue-43572-unused-uses test case as a result of #44435 (diff).

Maybe not a big deal, but worth doing a quick profile to see if something can be done!

cc @alexcrichton

[alexcrichton]

graph

[alexcrichton]

Looking at a profile the numbers are pretty close together, but I think it basically says that we're hashing more in blake2b

[alexcrichton]

I'm not 100% sure, but #44584 may be the culprit here

[nikomatsakis]

The fix that @alexcrichton landed improved things, as you can see from this graph, but it's not completely resolved.

[nikomatsakis]

triage: P-high

Calling this P-high as we investigate.