Add support for ISO week dates

[movermeyer]

What are you trying to accomplish?

Partly addresses #131.

What approach did you choose and why?

• Copied the helper methods from cPython source into a new isocalendar.c

• Added the two new branches, by porting the upstream code

• Updated the automatic test generation code:

  • to support these cases
  • to also return the reason that the timestamp is invalid. Useful for debugging.

I also had to exempt iso_week from the "too few characters" invalid test generation for cases where it was directly followed by iso_day. If you reduce the iso_week field to 1 character, then the following iso_day will make it into YYYYWwD which gets considered to be a valid YYYYWww timestamp. We don't have this problem with non-week-date timestamps, since YYYYMM is not a valid timestamp format, and day is always two characters (so the equivalent YYYYMD cannot happen). Note that this isn't a problem change the parser, but the way the invalid test case generation code was working.

• Updated the docs since we now support the same subset as Python itself

What should reviewers focus on?

Performance for non-week-date timestamps is imperceptibly impacted.

The impact of these changes

ciso8601 will now support the same subset of ISO 8601 as Python itself.

[AlecRosenbaum]

looks good so far, but please re-request review once tests are passing.

[movermeyer]

@AlecRosenbaum Sorry about the noise. I realized that I was missing the valid case of YYYYWwwD.

[movermeyer]

This is a reStructured text hack to allow links on inline code blocks.