Filtering for train/val/test splits

[RemyLau]

https://github.com/krishnanlab/NetworkLearningEval/blob/1a75b5b1e4525dce81e261d33d13c151cf595135/src/NLEval/label/LabelsetCollection.py#L344-L351

There is a potential issue with the current scheme that would lead to unnecessary removal of label sets. For example, in an iteration, multiple labelsets are to be removed by the LabelsetRangeFilterTrainTestPos filter, but some of them might be okay after reassigning the train/val/test splits.

One potential solution would be reassigning train/val/test splits after every removal of a labelset. However, this poses potential issues such as slower runtime, and also the ordering of removal, which might also lead to different final solution.

[RemyLau]

The optimal solution (for removing the least amount of data to make the split fulfilling the criterion) might not be trivial to obtain. Also given that SplitLSC is deprecated soon, see #72, this won't be needed anyways.