Unpin scipy

[larsevj]

With scipy version 1.15.0 the following test fails: https://github.com/equinor/ert/actions/runs/12628338741/job/35184324189

Seems to be a difference in output from the fcluster method https://docs.scipy.org/doc/scipy/reference/generated/scipy.cluster.hierarchy.fcluster.html

Running the following script with scipy 1.14.1 produces: [2, 1, 1], whereas with scipy 1.15.0 you get [3,1,2] .

import numpy as np
from scipy.cluster.hierarchy import fcluster, linkage

C = np.array([[1.0, 0.2, -0.2], [0.2, 1.0, 0.3], [-0.2, 0.3, 1.0]])
Z = linkage(C, "average", "euclidean")
print(fcluster(Z, 0, criterion="maxclust", depth=2))

[xjules]

Relates to #9672

[xjules]

Can we do a test how stable the clusterings are when using new scipy compared to the old one?