Avoid matches on pure stop words

[lonvia]

Photon is pretty dump in handling names and will happily match on any part of the name including stop words like articles, generics like 'street' or 'road'. This leads to really surprising results at times.

Two major tasks here:

• Implement an algorithm to identify stop words. This can be tricky because it is highly language- and location dependent.
• Change the index to ensure that the presence of stop words only improves scoring but doesn't cause a match.

Also to take into account: we cannot work with a constant list of stop words. What is a stop word in one name might be a relevant piece of information in another name. In the same language and location.

[hiSandog]

The hard part seems to be making stop words reduce match eligibility without deleting useful local tokens. One approach might be to keep stop-word tokens indexed but mark them as low-value terms, so they can boost an already relevant result but cannot satisfy a query by themselves. That would also allow language/location-specific stop-word learning later without breaking names where a word that looks generic in one region is actually distinctive in another.