Switch to SQLAlchemy Core for OSM objects

osm2pgsql creates tables without primary keys. Some OSM objects are
split into multiple table rows to help with PostGIS index performance.

Adjust the code to be aware of this.

Also add name matching for streets.

web_view.py

@@ -343,17 +343,54 @@ def api_get_item_tags(item_id):
                         duration=t1)
 
 
+def expand_street_name(name):
+    ret = {name}
+    if any(name.startswith(st) for st in ('St ', 'St. ')):
+        first_space = name.find(' ')
+        ret.add("Saint" + name[first_space:])
+
+    if ', ' in name:
+        for n in set(ret):
+            comma = n.find(", ")
+            ret.add(name[:comma])
+    elif '/' in name:
+        for n in set(ret):
+            ret.extend(part.strip() for part in n.split("/"))
+
+    return ret
+
+
 @app.route("/api/1/item/Q<int:item_id>/candidates")
 def api_find_osm_candidates(item_id):
     t0 = time()
-    bounds = read_bounds_param()
     item = model.Item.query.get(item_id)
     if not item:
         return cors_jsonify(success=True,
                             qid=f'Q{item_id}',
                             error="item doesn't exist")
 
-    nearby = api.find_osm_candidates(item, bounds)
+    label = item.label()
+    item_is_street = item.is_street()
+
+    if item_is_street:
+        max_distance = 2_000
+        limit = None
+        names = expand_street_name(label)
+    else:
+        max_distance = 400
+        limit = 60
+        names = None
+    nearby = api.find_osm_candidates(item,
+                                     limit=limit,
+                                     max_distance=max_distance,
+                                     names=names)
+
+    if item_is_street and not nearby:
+        # nearby = [osm for osm in nearby if street_name_match(label, osm)]
+
+        # try again without name filter
+        nearby = api.find_osm_candidates(item, limit=100,
+                                         max_distance=1_000)
 
     t1 = time() - t0
     return cors_jsonify(success=True, qid=item.qid, nearby=nearby, duration=t1)