Should choose ms/ns granularity for timestamp comparisons on-the-fly

[pnkfelix]

Spawned from #13 (comment)

Before running tango at all, do a local test of the host system to see if one can record nanosecond level deviations in the timestamps. (E.g. read the tango.stamp timestamp, add 1 nanosecond to it, re-stamp the file with that incremented value, re-read the new stamp, and see if its different from the value we first read.)

• If we can record nanosecond level precision, then use nanosecond granularity when comparing timestamps of files.
• If we can't record nanosecond level precision, then use millisecond granularity when comparing timestamps of files.

IMO it is not worth attempting to do this until after alexcrichton/filetime#9 is fixed.