coercions perf benchmark regressed by 3% to 14% after bidi lint addition

[pnkfelix]

Spawned off of #90462 (comment)

A bunch of benchmarks regressed. Its understandable to see overhead from a new on-by-default lint, especially one that operates at the lexing phase (and thus will affect a very broad range of benchmarks).

But coercions seems like it was hit particularly hard in some cases. I have to imagine we could do something to address that 14% regression to coercions check. (Note though that the biggest regressions are only on incremental builds of coercions. Full builds of coercions regressed by "only" 6.7%.)

[hkratz]

@rustbot claim

[hkratz]

#90559 fixes most of the performance regression so I think this issue can be closed.

[alex]

a) It sounds like this can be closed
b) I believe it should be I-comptime, not I-slow