ENH use Scipy isotonic_regression

[lorentzenchr]

Reference Issues/PRs

Closes #27378.

What does this implement/fix? Explain your changes.

Use scipy.optimize.isotonic_regression as of scipy 1.12 instead of our own _inplace_contiguous_isotonic_regression.

Any other comments?

[github-actions]

✔️ Linting Passed

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

_{Generated for commit: 4ca6e8b. Link to the linter CI: here}

[lorentzenchr]

Within uncertainty, this shows no performance regression:

This PR

% python benchmarks/bench_isotonic.py --iterations 10 --log_min_problem_size 2 --log_max_problem_size 8 --dataset perturbed_logarithm

100 0.0004406755011586938
1000 0.00043975089865853076
10000 0.00060384689932107
100000 0.0022764566019759513
1000000 0.02123089429878746
10000000 0.1800829847015848
% python benchmarks/bench_isotonic.py --iterations 10 --log_min_problem_size 2 --log_max_problem_size 8 --dataset pathological
100 0.0004570792989397887
1000 0.0004778172013175208
10000 0.0008545864999177866
100000 0.005397420399094699
1000000 0.06050054539737175
10000000 1.0579413614992519

MAIN

% python benchmarks/bench_isotonic.py --iterations 10 --log_min_problem_size 2 --log_max_problem_size 8 --dataset perturbed_logarithm
100 0.00048800000000000004
1000 0.0004021
10000 0.0005796000000000001
100000 0.0023363999999999998
1000000 0.019411199999999997
10000000 0.1722571
% python benchmarks/bench_isotonic.py --iterations 10 --log_min_problem_size 2 --log_max_problem_size 8 --dataset pathological
100 0.0004615
1000 0.00044409999999999995
10000 0.0008770999999999999
100000 0.005215
1000000 0.05959310000000001
10000000 0.9408343

[thomasjpfan]

Typically, we add this type of branching in utils.fixes. Although, given the scope, I am okay with this approach.

Nit: It's a little bitter to compare base versions:

Suggested change

	if sp_version >= parse_version("1.12.0"):
	if sp_base_version >= parse_version("1.12.0"):

Looks like related tests are failing

[thomasjpfan]

The CI failures look related to isotonic_regression.