Support converting from timedelta strings that lack seconds-place part

[LswaN58]

Currently, the logic that manages timedelta parsing doesn't handle timedeltas with missing seconds place. I'm starting to realize (nearly) the whole thing could probably be superseded by using dateutil.parser.parse(...), which as it turns out, is capable of parsing strings with timedelta-style formats. The only thing parse(...) doesn't handle is alphabetic characters that aren't specifying a date, e.g. "1 day, 5:00:00" would not parse.

When you parse a timedelta-style string, it comes back as a time attached to current day. Unlike pandas, it can handle times with out seconds, so we should include it in the list of parsers to try for the timedelta stuff. Between all the parsers, we'll eventually manage to handle our various cases.