Validation logic is inconsistent across languages

[willbeason]

There are two instances I found where validation logic for codes is inconsistent across languages:

1. Automatically expanding requested short code length if odd. Most of the
   language implementations automatically add 1 to the length of requested codes
   if the length indicates it is in the pair part of the code. Java, C++, Python,
   and Javascript are the outliers.

2. Validate latitude and longitude bounds within validation logic for full
   codes. Most of the language implementations only check the first two characters
   for latitude/longitude bounds when within the validation logic for full codes,
   not when generally checking if codes are valid. This does introduce a strange
   case where a code is "valid", but is neither a valid short nor a valid long
   code. C++ is the outlier here.

I have a PR out to fix this:

#750