Consider using Rationals for clock valuations

[morxa]

We currently use double for clock valuations, which already caused some bugs, e.g., #88.

We might want to use Rationals altogether. We should check again whether this would make sense.

[modass]

I believe this makes perfect sense, but we have to be aware that operations on rationals are about 200x slower (from experience, using libgmpxx).

Another option is, once we have fixed the granularity to use integers as representants (i.e., decide on a common divisor and only argue about and store the numerator) - this way we get the best of both worlds.