perf: Use Hashbrown for array_distinct

[neilconway]

Which issue does this PR close?

N/A

Rationale for this change

#20364 recently optimized array_distinct to use batched row conversion. As part of that PR, std::HashSet was used. This PR just replaces std::HashSet with hashbrown::HashSet, which measurably improves performance.

What changes are included in this PR?

Are these changes tested?

Yes.

Are there any user-facing changes?

No.

[neilconway]

Benchmarks:

  ┌────────────────────┬──────────────┬───────────┬─────────────┐
  │     Benchmark      │ std::HashSet │ hashbrown │ Improvement │
  ├────────────────────┼──────────────┼───────────┼─────────────┤
  │ high_duplicate/10  │ 145.00 µs    │ 68.94 µs  │ -52.4%      │
  ├────────────────────┼──────────────┼───────────┼─────────────┤
  │ high_duplicate/50  │ 719.46 µs    │ 350.68 µs │ -51.3%      │
  ├────────────────────┼──────────────┼───────────┼─────────────┤
  │ high_duplicate/100 │ 1.404 ms     │ 674.83 µs │ -52.0%      │
  ├────────────────────┼──────────────┼───────────┼─────────────┤
  │ low_duplicate/10   │ 180.33 µs    │ 107.87 µs │ -40.0%      │
  ├────────────────────┼──────────────┼───────────┼─────────────┤
  │ low_duplicate/50   │ 849.26 µs    │ 530.89 µs │ -37.6%      │
  ├────────────────────┼──────────────┼───────────┼─────────────┤
  │ low_duplicate/100  │ 1.706 ms     │ 987.72 µs │ -42.0%      │
  └────────────────────┴──────────────┴───────────┴─────────────┘

[comphead]

Thanks @neilconway it is amazing how 1 line can change so much

[neilconway]

Thanks @neilconway it is amazing how 1 line can change so much

Indeed; the choice of hash function in std seems a bit silly, at least for our use-case. #19869 might be worth considering...