Fix scipy<1.15.0 pin in virtual environment configuration files once issue in SciPy is resolved

[vschaffn]

Description

The latest release of SciPy (v1.15.0), published on January 3, 2025, causes errors in xdem during the execution of test_biascorr.py. Specifically, the following tests are impacted : test_biascorr__bin_and_fit_1d and test_biascorr__bin_and_fit_2d.

As a temporary fix, the scipy version has been pinned to <1.15.0 in environment.yml, dev-environment.yml and requirements.txt (#675) to prevent these errors from blocking development and CI.

To-Do

• Monitor updates from SciPy regarding this issue.
• Once the issue is fixed in a future SciPy release, remove the version pin and update the scipy dependency in environment.yml and dev-environment.yml to allow newer versions.
• Re-run the tests to verify that the issue is resolved in xdem.

Note

requirements.txt will be automatically synchronized with environment.yml via a pre-commit hook.

[rhugonnet]

Here are the error messages:

FAILED tests/test_coreg/test_biascorr.py::TestBiasCorr::test_biascorr__bin_and_fit_1d[nanmean-10-curve_fit-<lambda>-fit_args0] - RuntimeError: Optimal parameters not found: gtol=0.000000 is too small, func(x) is ort...
FAILED tests/test_coreg/test_biascorr.py::TestBiasCorr::test_biascorr__bin_and_fit_2d[nanmean-bin_sizes1-curve_fit-<lambda>-fit_args1] - scipy.optimize._optimize.OptimizeWarning: Covariance of the parameters could not be es...
FAILED tests/test_coreg/test_biascorr.py::TestBiasCorr::test_biascorr__bin_and_fit_2d[nanmean-bin_sizes1-curve_fit-<lambda>-fit_args2] - scipy.optimize._optimize.OptimizeWarning: Covariance of the parameters could not be es...

Two warnings and one error during scipy.optimize.curve_fit().

From past experience with another SciPy optimize error (1.12 update broke the behaviour of curve_fit with a lower bound of zero, which we had in SciKit-GStat/xDEM), they would have much more opened issues on their GitHub right now (after 2 weeks) if this was a common.
As they don't, this is a probably just a slight change in covariance estimation of the algorithm, which only affects some of our bias-correction tests that are quite sensitive as they have very low point number on purpose (to increase testing speed).

We can simply silence the warnings for the bin_and_fit_2d, as this is common.
But I'll look more into the RunTimeError...