Enhancement/switch multiprocessing to multithreading

[tgrandje]

This is a suggestion to switch from multiprocessing to multithreading for the geodata package:

• there are sometimes troubles using multiprocessing with windows machines, which are not easily debugged;
• as far as I have seen, there is not much interest for multiprocessing as no heavy computation is done, and (I think) we're perfectly safe with multithreading.

This needed to rewrite the creation of the session object, which is now created outside the pool scope (instead of global objects created at threads/processes creation).

[tgrandje]

This is derived from this PR. As the geoparquet issue is still waiting for concensus, I chose to open a new PR working on this particular subject instead.

As already said here, the choice of using pebble is mostly lazyness. The advantage is that it's API si very similar to that of multiprocessing (instead of something like concurrent.futures for instead). I've never had any trouble installing pebble, so I think it is reliable enough for this...