MNT: bump to sklearn 1.9 and adapt to it

[cakedev0]

Checklist

• pytest passes locally, with sklearn 1.9.0rc1 installed manually.
• code is well formatted
• documentation renders correctly
• CI passes

What does this implement/fix? Explain your changes

Bump scikit-learn to 1.9.0, and update the code to adapt to some changes in scikit-learn:

• remove criterion param from GradientBoostingSurvivalAnalysis, see PR MNT: trees/forests/GBT: deprecate "friedman_mse" criterion scikit-learn/scikit-learn#32708
  • I remove it abruptly, but maybe you want to follow depreciation cycles like we do in scikit-learn?
• stop handling missing values in the LogrankCriterion, this is now handled entirely outside of criterion logic in scikit-learn. See PR FEA Add support for missing values in tree estimators with criterion="absolute_error" by greatly simplifying the logic scikit-learn/scikit-learn#32119
• Stop importing DTYPE from sklearn.tree._tree, use np.float32 directly instead (which reflects what was done in scikit-learn)
• sample weights: call _check_sample_weight to align on scikit-learn
• sample weights in forests: pass them to scikit-learn bootstrap / OOB helpers, and adjust max_samples validation accordingly
• tests: update expected scores, I think these changes are mostly due to slight numerical behavior changes leading to different tie breaking. But I haven't check/bisect thoroughly, it would take a looot of time, even for Codex.
• tests: use CalibratedClassifierCV around SVC because SVC(probability=True) is now deprecated

[codacy-production]

Up to standards ✅

🟢 Issues 0 issues

Results:
0 new issues

View in Codacy

🟢 Metrics 5 complexity · 0 duplication

Metric	Results
Complexity	5
Duplication	0

View in Codacy

_{NEW Get contextual insights on your PRs based on Codacy's metrics, along with PR and Jira context, without leaving GitHub. Enable AI reviewer
TIP This summary will be updated as you push new changes.}

[cakedev0]

It seems the score improved! 🎉

But according to Codex:

I’d frame this as a numerically fragile tie exposed by sklearn tree splitter changes, not a meaningful model improvement and not directly the friedman_mse GB bugfix.

The difference only happens at iteration 88 of the GB due to an improvement delta of 3.6e-15 between two options. This is pure chance.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 98.46%. Comparing base (0361ab9) to head (ba407d7).

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #593   +/-   ##
=======================================
  Coverage   98.46%   98.46%           
=======================================
  Files          38       38           
  Lines        3715     3723    +8     
  Branches      481      484    +3     
=======================================
+ Hits         3658     3666    +8     
  Misses         27       27           
  Partials       30       30           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.