Support inter-run caching

[ssokolow]

In order to streamline the "find dupes, do something manual, re-determine dupes" workflow that often emerges in one-off situations, there should be some way to cache data between runs.

(Something like a --hash-cache=/path/to/file option.)

I'm thinking probably a key-value store which maps inode values to (size, ctime, sha1) triples so that fastdupes can take an rsync-inspired "assume that inode+size+ctime = sha1" shortcut.

(and probably at least starting out SQLite-backed given my bad experiences with shelve corruption, the need for incremental update support, and my aversion to premature optimization.)