Update numpy, scipy, scikit-learn deps

[markotoplak]

No description provided.

[markotoplak]

test_run_with_sparse_matrix_ignores_normalization and test_normalized_gives_correct_result fails are there because biolab/orange3#6815. These are the only issues in the non-conda Windows Installer and x86 Mac. These issues also appear on -oldest tests on orange3.

Arm Macs have additional issues: Orange.tests.test_random_forest.RandomForestTest.test_max_features_reg and Orange.widgets.evaluate.tests.test_owpermutationplot.TestOWPermutationPlot.test_info, which also fail on MacOS tests on orange3.

Windows conda release has an unknown issue though:

======================================================================
FAIL: test_expression (Orange.widgets.model.tests.test_owcurvefit.TestOWCurveFit.test_expression)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "D:\test-install\Lib\site-packages\Orange\widgets\model\tests\test_owcurvefit.py", line 373, in test_expression
    self.assertIsNotNone(model)
AssertionError: unexpectedly None

[markotoplak]

Test expression is not a problem in releases/3.37.0, which is on scikit-learn 1.3, so maybe 1.4 is the problem.

[markotoplak]

Conda installer is not tested here because new deps are not compatible with the released Orange version.

[markotoplak]

This updates scikit-learn to the required version, scipy to 1.13, before some sparse deprecations (orange3 works perfectly with 1.14 too, just keeping safe for addons), numpy to 1.26 (out for more than a year now).