Use OSM data for geocoding in all boroughs

[danvk]

#174

data update: oldnyc/oldnyc.github.io@171c2fb

Changed: 532
  +geom: 638

So far as I can tell, the additions are all good. The changes are mostly good. Of the 65 biggest movers, 5 are losses, 3 are neutral (neither is right) and 57 are wins. Pretty good! Most of these are correcting Google failures or previous bugs in the grid geocoder. The truth data supports this, we pick up ~20% of the remaining locations with zero losses:

-    208 / 269 = 77.32% of locatable images correctly located.
+    221 / 269 = 82.16% of locatable images correctly located.
-      7 / 215 = 3.26% incorrectly located.
+      7 / 228 = 3.07% incorrectly located.

This dramatically reduces reliance on Google for geocoding. I've also removed most all the per-item logging, at least for now. The net effect is that geocode.py is pretty fast! Only ~5s to geocode the ~41k entries in images.ndjson.

TODO:

• Generate all intersections data files (or remove some of them)
• Purge geocache
• Fix the "b" images in images.ndjson: "734792b", "716725b", maybe others (follow-up: Remove backing images from images.ndjson #181)
• Since this is a big change anyway, is this the right time to standardize rounding from Locatable refactor #177?