Prefilter input features

[shraddhapai]

A prefiltering module that reduces the number of input features that go into the model and improve memory footprint of predictor. Would make use of gene-level or unit-level features feasible.
It could do the following:

• canned unit-wise t-tests (e.g. limma) that prune out units that aren't distinguishable among groups (e.g. pruning RNA datapoints after diffEx test)
• create meta-features such as eigen-genes by PCA or shrinking centroids
• pruning genotypes by GWAS pvalue.
• pruning genotypes by excluding those not associated with a GSEA signal.

[shraddhapai]

runPredictor_nestedCV seems to do this already. Check that it works by writing a use case.