Optimize compute_live_ins in optimizer hotpath

[LilithHafner]

This is my first time touching the compiler, so I'd appreciate a careful review!

I'm assuming that defs and uses are disjoint and increasing and that block_for_inst is a nondecreasing function.

This is motivated by my belief that almost all uses of sort can be replaced with more performant alternatives.

[aviatesk]

@nanosoldier runbenchmarks("inference", vs=":master")

[nanosoldier]

Your benchmark job has completed - no performance regressions were detected. A full report can be found here.

[JeffBezanson]

Any noticeable compile time improvements from this?

[LilithHafner]

None that I know of. Perhaps this was not really a bottleneck.

[thofma]

A bisect shows that this broke Hecke. Here is an example for the backtrace: https://s3.amazonaws.com/julialang-reports/nanosoldier/pkgeval/by_date/2022-09/22/Hecke.primary.log
(Do not be deceived by the "successful" at the end.)

Should I open a separate issue?