Warn if we find a Unicode homoglyph for a symbol in an identifier.

Specifically, warn if:
 * we find a character that the language standard says we must treat as an
   identifier, and
 * that character is not reasonably an identifier character (it's a punctuation
   character or similar), and 
 * it renders identically to a valid non-identifier character in common
   fixed-width fonts.

Some tools "helpfully" substitute the surprising characters for the expected
characters, and replacing semicolons with Greek question marks is a common
"prank".

llvm-svn: 320697