Annoy index files are standalone and can be used by anyone to perform similarity lookups.

We should make these files available to users as downloads so that they can use them if they want to do bulk lookups. This means that we'll also have to provide a mapping of ll.id -> mbid/offset, as the ll.id is used as the index in annoy.

Consider also providing a dump of the similarity.similarity table for people who want to do their own non-annoy similarity