Fixed bug in TimeOffset.to_count (both versions) that caused incorrect results when rounding down when the denominator of the rate properly divided neither the seconds part nor the nanoseconds part of the timestamp, nor the number of nanoseconds in a second, but did properly divide the whole timestamp expressed in nanoseconds. This bug has been present for some time but had not previously been seen because there was no test coverage for non-nearest rounding and none of our tests used rates with prime denominators other than 2.
Switched test runner in tox from nose2 to unittest