Extend GMM-MI to work with multivariate random variables, too

[dpiras]

As explained in #17

[dpiras]

We will wait to merge until we test this new feature against the results in Sui et al. (2023), as explained in #17. We will then update the Pypi version to 0.9.0, too. It might be useful to add a warning sating that these high-dimensional capabilities are not still fully tested.

[dpiras]

For now, we observe that when applying our estimator on data from Sui et al. (2023), we get consistent results, but crucially with the PS&BS estimates below the PS estimates for the pure signal and the thermal noise + foreground cases. We suspect this is due to either outliers in the PS features, or to the suboptimal choise of hyperparameters when fitting GMMs in such high dimensions (>50); however, so far we found no quick way to fix this. It might require more work, so for now we will search for other datasets to test the high-dimensional MI, and if possible we will test more hyperparameter configurations.