Forgot to return after finding our first, largest blob during blob
tracking, meaning we'd possibly return many blobs even though
we can't do anything with that info.
Fixes#20
Use a python event as a oneshot channel to request the capture source
to get a frame, versus letting it poll and serve at its own speed
(which will murder us on file loads)
Divide out utilities from main algorithm, set utilities on their own
threads. Reference binaries in original directory so we don't have to
duplicate them in the repo.