Reduce code duplication in find* functions

[goretkin]

Try to address #36963

Also, whatever fix arises from #36938 will only need to be applied in a fewer number of places.

I expect the compiler can certainly specialize on identity. Is it necessary to force inlining?

[goretkin]

predicate and non-predicate versions were introduced by #25577

[nalimilan]

Sounds good, but can you run some benchmarks and look at the generated code to check that this doesn't introduce regressions?

[goretkin]

Yeah, it would be good to have a sense that there aren't performance regressions. Can nanosoldier help here?

[nalimilan]

Yes, if there are tests in BaseBenchmarks for this (haven't checked) or if you make a PR to add them. But a quick local check would be enough IMHO. You can also check that the generated code is the same (which is likely).

[goretkin]

Looks like there are some: https://github.com/JuliaCI/BaseBenchmarks.jl/blob/ce11d8badcb0e9f6d8b0518ffd57a472128af930/src/find/FindBenchmarks.jl#L39

but I'll take a look at the generated code, since you're probably right that it didn't change.

[goretkin]

FWIW, I wanted a way to programmatically compare the generated code, and I didn't know how to do that since all the tools seem to print to stdout directly. Perfect was the enemy of good, but I'm glad to see this merged!