ENH sort before binning in HGBT

[lorentzenchr]

Reference Issues/PRs

Suggested in #28064 (comment).

What does this implement/fix? Explain your changes.

HGBT's _BinMapper searches for bin thresholds. This calls np.unique and then percentile which both internally use np.sort.
This PR first applies np.sort such that the subsequent internal calls to np.sort are applied on an already sorted array.

Any other comments?

The binning part of HGBT is very small compared to the actual fitting of trees (5s out of 60s for Higgs with 100 trees).

[github-actions]

✔️ Linting Passed

All linting checks passed. Your pull request is in excellent shape! ☀️

_{Generated for commit: 5aebdcd. Link to the linter CI: here}

[adrinjalali]

LGTM.

Especially since:

In [1]: import numpy as np

In [2]: a = np.random.rand(10_000_000)

In [3]: %timeit np.unique(a)
847 ms ± 7.53 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)

In [4]: b = np.sort(a)

In [5]: %timeit np.unique(b)
177 ms ± 1.19 ms per loop (mean ± std. dev. of 7 runs, 10 loops each)