The procedure to convert a given time object to a float containingthe "seconds until the next run" was extracted into an auxiliarytop-level function named `_to_interval_seconds`.Also, `_put`'s "prep phase" was refactored as follows. A new localidentifier named `interval` was created to make its meaning clearer(instead of always reusing `next_t`, which was confusing). Theway in which the interval is fetched (and checked for `None`)has also been simplified. Finally, `next_t`'s last assignment isthe actual timestamp in which the job is to be run, which iseither `previous_t` (if available) or the current time plusthe amount stored in `interval`.Finally, `last_t` was renamed to `previous_t` to avoid confusionwith upcoming `last` kwarg.

Added a `last` kwarg to `run_repeating` that specifies whenthe job should last be run. The checking is done before executingthe job, but with a small delay buffer, so that if `last`is an exact multiple of `interval`, the last job is stillrun, even though the actual time is slightly after thetheoretical finish time.

Instead of converting to interval first and then addingthe current (or previous) time.

Python 2's datetime module doesn't have the datetime.timestamp()method. An equivalent calculation has been put in its place.