Implement full to_{upper,lower}case algorithms

[alexcrichton]

Right now we always return an iterator over one character, but the iterator is being returned so one day we can return many characters. This has all yet to be implemented, and this issue will track this implementation.

[andrewclarkson]

A couple friends and I dug into this issue as part of a mini-sprint. We'd love a mentor or some guidance to work on this.

As we understand it right now:

• Unicode case conversions are done using raw text tables from unicode.org
• The raw tables are exported as rust slices in tables.rs using a python script unicode.py
• The current to_uppercase methods use a binary search over the single character conversion tables ignoring the multiple character compatibility_table

So what needs to happen then is:

• Either join the two tables or search both of them.
• to_upper and to_lower in tables.rs then need to be changed to return a slice rather than a single char
• to_uppercase and to_lowercase need to use that slice for the iterators it returns

Is that an accurate summary?

Also, we noted there are several external but related crates for unicode but couldn't find any indication on whether these crates were moving into the standard lib or whether the standard lib was moving out.

[alexcrichton]

@bitborn that all sounds pretty good! I think we may not want to encode the exact return value of each character (that's a lot of space). It'll be a balancing act to figure out how to encode the data on unicode.org in as compact a form as possible but still having a fast lookup for case conversions.