dereplicate proteins on genus level to reduce database size

[jvollme]

maybe use 98% aminoacid identity cut-off?
proteins that are unique for one species in a genus would still be attributed to that individual species (but only one copy would be kept, in case of multi-copy entries)
"redundant" proteins, that occur identically in multiple species of a genus would be attributed to the genus instead of the species (again only represented by one copy)
--> reduces dataset size
--> increases diamond/blast search speeds
--> increases speed of LCA-classifications (a little bit)?