Use Hypothesis to find more bugs in the time parsing

[schwa]

Given the bug in #131 and another one found while working on [93d4fddb80c3], I can imagine that there are even more bugs. To find them, the ​Hypothesis library may be useful.

However, this has the problem that the parser code currently uses code from datetime and time, which implicitly use information on current time and my local time zone. To make Hypothesis effective, it would require a redesign of the tests and possibly the production code to be able to vary the "perceived" time and time zone.

So the use of Hypothesis would require a tradeoff between (hopefully) fewer bugs and a more complicated design.