Gracefully handle interrupted renders without requiring a --check-tiles run

[stwalkerster]

My thinking here is to dump the dirty tile list, the render start time, and a hash of the config to disk at the start of a render, then if/when a render is cancelled it can check mtime of region files vs mtime of all tiles in the dirty list, and add the old dirty list to whatever dirty list exists for the newly-started render? I'm just throwing ideas around at the moment - dumping a dirty tile count is definitely going to be easier than what I just suggested.

Originally posted by @stwalkerster in #69 (comment)