Replace Vector with BitSet in optimization passes

[LilithHafner]

I was unable to determine why this sorting operation is needed. I tracked it down to 8af40ea so perhaps Keno (waiting for CI to pass to ping) knows if there is a reason to keep this?

Closes #46507

[vtjnash]

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

[nanosoldier]

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

[LilithHafner]

While the performance improvements are negligible, it seems plausible that the previous approach could result in quadratic runtime in certain cases due to

julia/base/compiler/ssair/passes.jl

Lines 78 to 80 in e1139e7

	while !(curblock in allblocks) && curblock !== 0
	curblock = domtree.idoms_bb[curblock]
	end

I think this change is worth making. Whitespace is broken due to old master.

[codecov]

Codecov Report

Merging #46587 (e1139e7) into master (c0d5ffb) will increase coverage by 0.05%.
The diff coverage is n/a.

❗ Current head e1139e7 differs from pull request most recent head 9eea1b8. Consider uploading reports for the commit 9eea1b8 to get more accurate results

@@            Coverage Diff             @@
##           master   #46587      +/-   ##
==========================================
+ Coverage   93.63%   93.69%   +0.05%     
==========================================
  Files         380      387       +7     
  Lines       85981    86056      +75     
==========================================
+ Hits        80508    80628     +120     
+ Misses       5473     5428      -45     

Impacted Files	Coverage Δ
compiler/ssair/passes.jl	85.50% <0.00%> (-5.35%)	⬇️
uuid.jl	97.10% <0.00%> (-2.90%)	⬇️
Base.jl	95.23% <0.00%> (-2.49%)	⬇️
env.jl	92.30% <0.00%> (-1.93%)	⬇️
compiler/ssair/ir.jl	79.72% <0.00%> (-0.85%)	⬇️
compiler/ssair/irinterp.jl	85.99% <0.00%> (-0.81%)	⬇️
compiler/ssair/inlining.jl	95.22% <0.00%> (-0.59%)	⬇️
operators.jl	97.15% <0.00%> (-0.56%)	⬇️
compiler/ssair/domtree.jl	94.06% <0.00%> (-0.46%)	⬇️
expr.jl	94.40% <0.00%> (-0.35%)	⬇️
... and 430 more

Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here.

[LilithHafner]

Bump (c.f. #46679 (comment)). There doesn't seem to be much practical performance impact of this PR, but it is a theoretical improvement and more importantly, it reduces Core's dependency on Sort, which is needed for #46679.

[LilithHafner]

Rerunning benchmarks after rebase
@nanosoldier runbenchmarks("inference", vs=":master")

[nanosoldier]

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

[oscardssmith]

I'm merging this today sans objections.