Apply InitialOptimizations more consistently in sorting

[LilithHafner]

#47383 revised how initial optimizations (special cases for eltypes that are unions with missing and/or a concrete IEEEFloat or Bool and very small input) are implemented and dispatched to. I accidentally dropped the initial optimizations in that PR for cases where the user specifies alg=QuickSort, alg=MergeSort, or alg=InsertionSort. I also dropped those optimizations for partialsort and paritalsort!. Thanks @vtjnash for catching these regressions here

This PR applies InitialOptimizations to achieve better performance and avoid regressions.

julia> @btime sort(rand(50_000); alg=QuickSort);
  2.705 ms (2 allocations: 390.67 KiB) => 3.733 ms (4 allocations: 781.34 KiB) => 2.652 ms (4 allocations: 781.34 KiB)

julia> @btime sortperm(vcat([missing], rand(50_000)));
  6.221 ms (8 allocations: 1.19 MiB) => 6.785 ms (12 allocations: 1.57 MiB) => 3.525 ms (12 allocations: 1.57 MiB)

julia> VERSION
v"1.8.3" => v"1.10.0-DEV.164" => PR

Also fixes a dispatch bug in ::MissingOptimization and debugs code that was previously unreachable due to that dispatch bug.

[LilithHafner]

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

[nanosoldier]

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

[KristofferC]

@nanosoldier runbenchmarks("sort", vs=":release-1.8")

[nanosoldier]

Your benchmark job has completed - possible performance regressions were detected. A full report can be found here.

[KristofferC]

Performance now looks equal to 1.8 but allocates 3x more. Is that expected?

[LilithHafner]

The allocations in sortperm! should be zero but there is a type instability that caused about 16 bytes and 1 allocation in 1.8 and now causes 48 bytes and 3 allocations. The nanosoldier benchmark is basically this:

julia> list = rand(50_000);

julia> ix = collect(1:length(list));

julia> @benchmark sortperm!(x, $list; alg = InsertionSort, rev = true) setup=(x = copy($ix))
BenchmarkTools.Trial: 5 samples with 1 evaluation.
 Range (min … max):  1.068 s …   1.144 s  ┊ GC (min … max): 0.00% … 0.00%
 Time  (median):     1.087 s              ┊ GC (median):    0.00%
 Time  (mean ± σ):   1.091 s ± 30.354 ms  ┊ GC (mean ± σ):  0.00% ± 0.00%

  ▁  ▁          █                                         ▁  
  █▁▁█▁▁▁▁▁▁▁▁▁▁█▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁█ ▁
  1.07 s         Histogram: frequency by time        1.14 s <

 Memory estimate: 48 bytes, allocs estimate: 3.

See issue #47949

[LilithHafner]

The performance diff across this PR seems reasonable under JuliaCI/BaseBenchmarks.jl#305 as well

[KristofferC]

Alright. This should be good to go then?