Stop using unexported Base.permute!!

[LilithHafner]

This PR may increase performance in some cases and should pretty much never pose much of a regression. Perhaps more importantly, it drops a dependency on an undocumented internal method of Julia that may be removed in 1.10 because internals are not subject to SymVer.

In 1.9.0, Julia's permute! method got a bit more efficient. These benchmarks compare use of permute! to permute!!

julia> using BenchmarkTools, Random, Plots

julia> x = 2 .^ (3:26);

julia> GC.gc(); y!! = [(hmer_heights = rand(Int, n); o = randperm(n); @elapsed(Base.permute!!(hmer_heights, o))) for n in x];

julia> GC.gc(); y! = [(hmer_heights = rand(Int, n); o = randperm(n); @elapsed(permute!(hmer_heights, o))) for n in x];

julia> plot(x, y! ./ x, label="permute!"); plot!(x, y!! ./ x, label="permute!!", xlabel="length(hmer_heights)", ylabel="seconds/element", xscale=:log10, yscale=:log10)

In the case of the usage in this package, in cases where permute!! is faster (small inputs or old versions of Julia) this PR may result in a minor performance regression, but never more than 33% because there are already two calls to permute! in this spot. In cases where permute!! is slower, there is not really a cap on how much of an improvement this PR can be. For large inputs on modern versions of Julia, this could easily be a 3x or more improvement because of the poor cache locality properties of permute!!.

(I rewrote this post with the release of 1.9.0)

[codecov-commenter]

Codecov Report

Patch coverage: 100.00% and project coverage change: +0.14 🎉

Comparison is base (8af2507) 95.30% compared to head (ec4a764) 95.44%.

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #229      +/-   ##
==========================================
+ Coverage   95.30%   95.44%   +0.14%     
==========================================
  Files          17       18       +1     
  Lines        1213     1448     +235     
==========================================
+ Hits         1156     1382     +226     
- Misses         57       66       +9     

Impacted Files	Coverage Δ
src/hclust.jl	94.65% <100.00%> (-0.21%)	⬇️

... and 13 files with indirect coverage changes

☔ View full report in Codecov by Sentry.
📢 Do you have feedback about the report comment? Let us know in this issue.

[LilithHafner]

Bumping because now that Julia 1.9 is out, this is a pretty clear performance win in addition to a maintainability win.