Make generator to parse data for performance improvement with large files 

Since some `*.osm.pbf`  files are quite large, do you think it would be possible to `yield` each row of data instead of appending it to a `dict`? So that `parse_osm_pbf()` would become a generator. This would be great for my use-case, if you want I can start a PR for this feature if you think its useful/feasible. @mikeqfu 