improve unit number sorting for fallback.housenumber queries (/v1/search)

[missinglink]

Issue as described in pelias/pelias#810
ref: pelias/query#139

This PR resolves a long-standing issue with /v1/search where exact matches of unit numbers are not given priority in the sorting.

This is accomplished by adding a 'should' condition to all fallback.address queries which uses the original housenumber as parsed by libpostal (including both alpha an numeric tokens) to match against the phrase.default field, increasing its score slightly compared to the other matches.

The reason this is required is because the existing match_phrase address_parts.number query uses a tokenizer which strips non-numeric tokens.

resolves pelias/pelias#810

Some caveats which are worth noting but not an issue in practice:

• If the parse contains the housenumber 1 (for instance) and the document name.default contains that token in any position (ie. 8 1 street) then it will get a boost. This is likely not an issue as the address_parts.street must match.
• If the parse contains the housenumber 1 (for instance) and the document name.default contains that token multiple times (ie. 1/1 main street) then it will get multiple boosts. This is likely not an issue but worth keeping an eye on.
• If the parse contains the housenumber 1 F (for instance) and the document name.default contains the token 1F then it will get no boost. This is something we might be able to improve.

https://pelias.github.io/compare/#/v1/search?text=kinkerstraat+175F&debug=1