Move the first cutting of the result list before reranking
by result match. This means that results with significantly
less importance are removed early and independently of the
fact how well they match the original query.
Fixes#3266.
The algorithm is similar to the PHP reranking and uses the terms from
the display name to check against the query terms. However instead of
exact matching it uses a per-word-edit-distance, so that it is less
strict when it comes to mismatching accents or other one letter
differences.
Country names get a higher penalty because they don't receive a
penalty during token matching right now.
This will work badly with the legacy tokenizer. Given that it is
marked for removal, it is simply not worth optimising for it.
Search is now split into three functions: for free-text search,
for structured search and for search by category. Note that the
free-text search does not have as many hidden features like
coordinate search. Use the search parameters for that.
