Add BLAST-based identification wrapper

[brendanf]

Mothur's k-nearest-neighbors and CREST's LCAClassifier are both variations on "take a bunch of BLAST hits and use the LCA consensus". K-nearest-neighbors chooses a set number of hits to take for all queries, LCAClassifier chooses based on comparisons between BLAST alignment scores. Both of these, as well as the naïve "just take the best BLAST hit" method (i.e., k-nearest-neighbors, k = 1) could be implemented as methods for phylotax::taxonomy(), via an external call to megablast.