improved matching of street directional prefixes

[missinglink]

Looking over the code today I noticed that the query 'Am Falkplatz' (and friends) are relying on a StopWordClassification for matching.

This seems brittle, this PR replaces it with the more robust DirectionalClassification.

I've also removed am from the /de/stopwords.txt file as Am Falkplatz is a valid streetname in Germany, the preposition shouldn't be ignored.

[missinglink]

In fact, it seems as though Am Falkplatz wasn't even relying on the scheme/street.js classification which bore its name.

Removing that blocks only makes one test fail: N FISKE AVE Portlan

[missinglink]

Worth noting this regression I spotted which wasn't covered by tests:

node bin/cli.js 10 A Main Street

(0.98) ➜ [ { housenumber: '10' }, { street: 'A Main Street' } ]

After this PR there are no results returned.

The previous parse wasn't correct anyway, so I'm not sure where I stand on it 🤔

The derived PR #180 seems to fix it (the reason is the TokenDistanceFilter penalises it because the housenumber and street are too far apart).

[Joxit]

LGTM

[missinglink]

I merged this today, afterwards I confirmed that the query 10 A Main Street returns zero results (which wasn't covered by tests).

As I intended to merge #180 immediately afterwards, at that PR restores this query, and actually improves the parse, I'm hoping that it didn't break anything.

We need to keep an eye out on this one for any other regressions not covered by unit tests and consider reverting it, or fixing it, if other issues surface.