Provide fast fuzzy search

[frabert]

Either use ripgrep, or:

• generate an index of trigrams of each symbol at index time
• compute trigrams of the query
• use a MinHash scheme to find best candidates

this has the potential to be faster than ripgrep, but is limited to symbol names

[pgoodman]

I think one low/medium effort way to achieve this could be the following:

• [ ] Extract out the data for files from inside the rpc::File, and put it into a column of the file table.
• [ ] Add a fts5 virtual table that uses the file table as an external content table, where the rowid maps to file.file_id.
• [ ] Use snippet and this as an example of how to correlate matches with specific file offsets.
• [ ] Provide an api in the entity provider, implemented in the SQLiteEntityProvider, to get the file ids and byte ranges matching the search.
• [ ] Work it into an actual API, that yields out matching tokens.