Faster filtering

[kasper9n]

After moving to SQLite, we can store 1-3 ngrams in a table.

We store both a normalised version and an exact version. If query_term != normalised(query_term), then we only check exact ngrams for that term.

For example, with 3-grams:

`e.ee` gets stored as
`eee` (normalised)
`e.e`
`.ee`

`ééé` gets stored as
`eee` (normalised)
`ééé`

`one  two` gets stored as
`one`
`two`
`ne `
`e  `
`  t`
` tw`

Even though we store whitespace, we would only get whitespace results if the user does an "exact search"

CREATE TABLE search_ngrams (
    ngram    TEXT NOT NULL,
    field    INTEGER NOT NULL,
    is_normalised BOOL NOT NULL,
    track_id TEXT NOT NULL,
);

For 1-3 character query terms, we know that our results are correct. But for 4+ char query terms, we would have to verify them.

Problem: Would need a fast way to query this, which does not seem to exist