49 some binarized predictions are calculated using mtp threshold at level of country of interest instead of europe

[soriadelva]

This pull request addresses two main issues:

• It fixes the calculation of the MTP threshold for creating binarized (presence/absence predictions) when focussing on a specific country of interest. In this case, the MTP was calculated on the observations in that country but this should have been on global occurrences for the climate model and on European occurrences for the habitat model
• It fixes memory issues related to the PCA step for both models, by now calculating a PCA on 100000 randomly sampled pixels. Preliminary analyses on a single species showed that this number seems to be sufficient to provide stable results. More extensive analyses in the future need to further defend this choice.