failing on estimation is stupid

[Andrew Garner]

holland should not try to estimate the size of a backup and then fail. estimation may be useful for information purposes, but we should no longer fail on "oh no i may not have enough space".

Rather, holland should have a watchdog thread that monitors space and aborts cleanly if we exceed some configurable threshold. I imagine this working like pt-stalk's --disk-bytes-free, --disk-pct-free options, with perhaps somewhat larger defaults.

The actual implementation would periodically look at os.statvfs and then if we have exceeded the threshold, send a SIGTERM to the process group (so we can automatically terminate background commands) and then raise an appropriate BackupError in a signal handler.