Shortcut adding zero timedelta in parse_time()

[dstansby]

xref #5107 - there is a non-trival performance overhead in adding a zero timedelta, when instead one can just check for zero and not add the timedelta. This should possibly go upstream to astropy too?

[ayshih]

I recommend a complete change in the fix approach. time_delta can only have two values: TimeDelta(0*u.day) and TimeDelta(1*u.day). I suggest getting rid of time_delta entirely by changing the private function _regex_parse_time() to return a Boolean instead for its second argument, and that Boolean would drive if a day needs to be added in convert_time_str().

[dstansby]

Good idea, thanks for digging a bit deeper; modified as such.

[Cadair]

I think parse_time might be a great place to start writing some benchmarks?