feat: par_filter_map_collect

[claudioap]

I have a usecase that consists of bootstrapping a database with a subset of the planet. This subset is somewhat sparse and would not cause memory problems if it was to be eagerly allocated. I imagine other people to have similar requirements: basically being able to run a filter_map on the data.

One can do this in series with the supplied functions, however, not in parallel, which is a requirement for anyone who's doing frequent runs on big dumps.

I generalized my approach in this PR. Albeit more granular than what #42 is offering, this is slightly less elegant as it assumes that you want your blobs decoded and that you want to run through OsmData, which might not be true.

I could not get flatten() to let errors out and the workaround not a lazy iterator. This approach can easily fill someone's RAM if the filters happen to be too broad. Nonetheless, I've been reliably using it for a while now and just had the time to upstream it. Hope it serves someone else.

Cheers