add top_bottom filter to numerai_corr, add np.abs to max_feature_correlation

[cshanes]

two user reported issues:

• the top_bottom arg in numerai_corr doesn't do anything.
  • I wasn't 100% sure about whether or not it should happen before the tie_kept_rank__gaussianize__pow_1_5 step, but followed the same convention from signals diagnostics.
• the max_feature_correlation feature should have an abs because of negative correlations

These changes are not tested yet. @ndharasz lmk if this looks good, and if you aren't able to test I can do it and get this merged

[ndharasz]

Pretty sure this logic already exists in the pearson_correlation function. We just need to pass in top_bottom to that function.

[cshanes]

Pretty sure this logic already exists in the pearson_correlation function. We just need to pass in top_bottom to that function.

that was my initial thought, but wasn't sure if it mattered if top bottom filter happened before the tie_kept_rank__gaussianize__pow_1_5 step. if it doesn't matter then yea it should just be passed into pearson_correlation.

[ndharasz]

I was thinking that FNC did the tb200 filtering after neutralization so we should be doing tb200 filtering after any/all preprocessing