swift234

fix <rdar://problem/21196171> compiler should recover better from "unicode Specials" characters

There are a few changes here:
 - Exclude the "unicode specials" block (http://en.wikipedia.org/wiki/Specials_(Unicode_block))
   from being considered part of the valid identifier characters.
 - Have the compiler produce a fixit hint that replaces invalid UTF8 codepoints and invalid
   characters in general with whitespace with a blank space.
 - When the lexer sees an invalid character like this at the start of a token, rescan for the
   next token, improving error recovery, instead of returning tok::invalid.
 - Fix the order that diagnostics are generated in the curly quoted string case to report the
   end quote last.  This interacts better with Xcode's fixit hints, because just fixing the last
   quote allows the compiler to detect that remaining problem, but fixing the first quote hides
   the second one with our current implementation.



Swift SVN r29222