Calculate digests in parallel

A feature idea that @mostynb mentioned in his fantastic Bazelcon talk was for recc to calculate the digests in parallel. Currently we digest files sequentially, which leads to a bottleneck linear to the number of files we need to process.

For comparison, I believe goma calculates file digests in parallel and also maintains a cache. In the first instance, we could at least try and speed up the rate at which we calculate digests (within some specified limit).