findAnchorCells speed improvements

[patrickjdanaher]

• we should only calculate logliks for cells above the cosine distance threshold for at least one profile
• additionally: for any given cell, we only calculate logliks for the profiles with the 2-3 highest cosine distances

(The goal is to avoid calculating logliks for all cells * profiles, which is slow.)

Plan:

• calc cosine for all
• build a matrix of which cells will get their loglik calculated: 1. rows with at least one cosine above the threshold, 2. within those rows, only columns with cosines > 0.5 * threshold.

[patrickjdanaher]

Exact alg:

1. calc cosines
2. ID rows with max > thresh
3. For each cell type:
   • get subset of rows flagged in (2) AND with the cell type's cosine > 0.5*thresh
   • get lldist over that subset

[patrickjdanaher]

done.