I think since none of the >5 methods (see above) to record/derive street names on sidepaths is widely in use or accepted, only very specialized/experimental routing providers would have an incentive to use this kind of data.
IMHO this is a big gap in our data model, and closing it would have great potential.