Add ‘volume_stable’ invariant measure to improve numerical stability in NormalizedMI

[marko-tuononen]

This pull request introduces a new volume_stable invariant measure to the NormalizedMI estimator to enhance the numerical stability of mean k-NN volume calculations, particularly in high-dimensional spaces. The new measure is based on a logarithmic transformation technique that mitigates numerical overflow, as described in the manuscript “Improving Numerical Stability of Normalized Mutual Information Estimator on High Dimensions.”

Changes include:

• Implementation of the volume_stable invariant measure, providing a more stable calculation of the mean k-NN volume.
• Updates to the _scale_nearest_neighbor_distance function and INVMEASURES to support the new invariant measure.
• Expansion of unit tests to cover the volume_stable option, ensuring correctness across various metrics.

This addition improves the stability of normalized mutual information estimation, particularly in high-dimensional settings where standard volume normalization may lead to numerical instability (overflow).

[braniii]

@marko-tuononen Thx so much for your work and the MR. I will try to start the review asap.

[braniii]

@marko-tuononen Do you know why the pytest CI did not run on your branch/MR?

Nevermind. I've checked the code and it seems that there are (hopefully) no issues. Thx again for your contribution :)