Fix parsing of localized meridians (refers #228)

[swistakm]

changes

• added tests for meridians with and without localization
• a token now accepts only am/pm forms as stated previously
  in documentation
• A token accepts both am/pm and AM/PM forms to ensure backwards
  compatibility
• docs: updated output/accepted format of A & a tokens
• added footnotes to some tokens explaining localization

Additional notes:

token A: existing tests ensured that A token accepts both lowercase and uppercase meridians. I decided to keep this behaviour to ensure backwards compatibility. I think this could be changed in future to accept only uppercase but this would require bumping of major version part.

token a: I have decided to drop support for a/A/p/P values from following reason:

• single letter for instead of am/pm is a very rare case
• this was already documented as a => am/pm, A => AM/PM
• there is no support for single-letter meridians in locales.
• during formatting a already translates to am/pminstead of a/p/A/P and such inconsistency should be treated as a major issue. So this needed to be made consistent across format/parse and I think the best option is to follow already documented behaviour.

documentation: we have few tokens that supports localization. I have added two footnotes to tokens table because this was not obvious if dates with those tokens can be parsed using locale. Additionally the footnote on ddd and dddd says that these tokens support only formatting since we do not allow such tokens in parse calls.