shst geometries less reasonable than osmnx geometries

[yuqiww]

There are cases when one shst link ('shstReferenceId') contains multiple osmnx links ('osmid's or 'wayId's). When we aggregate osmnx ways back to shst links, we keep the attributes of the longest osmnx way. In many cases this is fine, but in some cases, the separate osmnx ways more accurately represent the roadway lane counts. In the example below, the highlight shst link has 'forwardReferenceId' '3043a16534961361d8c95b3754e38715', consisting of two osm ways: 392868123, 35836150. Using the length-based aggregation rule only keeps wayId 392868123 with lane =1 (imputed from "turn"), but this misses the small segment at the intersection that actually has 4 lanes, as shown in wayId 35836150.

@lmz : how critical is this type of misrepresentation of lanes?