Stricter parsing, support more phrases, use doctest

[exogen]

OK this one is intentional. :)

First off I know the point of this snippet is that it's tiny and simple. It's super handy and easy to modify. So you don't gotta accept this PR but I bet someone out there will find it useful.

My two big issues with the current snippet:

• Any combination of nonsense numbers are supported (they're just added together), e.g.:
  • text2num("one one") == 2
  • text2num("eleven three") == 14
  • text2num("thousand eighty hundred twelve hundred") == 801200
  • text2num("million three twenty") == 23
• Numbers with "implied hundreds" aren't parsed as one would expect, e.g.:
  • text2num("nineteen eighty four") == 103
  • text2num("one fifty") == 51

This PR fixes all those cases to do what you'd expect (throw an error or give a more sensible number). They could be guarded behind a strict flag if you wanna support both.

And some minor stuff:

• Adds support for "and" in appropriate places and , in the input string – this means we can round-trip output from the equally-handy num2words package. e.g. two thousand, one hundred and three.
• Adds a ton of new tests in doctest format in the docstring.

Thanks for the excellent snippet @ghewgill!