Improve hashing management for large file parts

[ppolewicz]

Currently during sync the main thread does all the hashing aggresively, which can put it hundreds or thousands of parts ahead of the uploading threads.

I suggest we introduce a new parameter (to emerge executor?) which will create a multiprocessing pool which will farm out those hashing requests to separate processes, so that it can be parallelized over many cores for a generally faster operation. This also partially addreses the pycurl problem where performance may increase to a point where we'll want to avoid hashing in the main thread.