All test pass but bufaccess, tnumpy, and r_mang1.

fixes the unicode literal indexing problem (only for unicode strings, not for byte strings!)

otherwise, different ways of spelling special characters can end up being correctly escaped or not in the C file

fixed unicode escape handling in byte strings
unescape \xXY in string literals as C allows it to conflict with trailing hex numbers - output string escaping will do the right thing

fix format string in __Pyx_AddTraceback() to let it match input types (line number is not defined as unsigned)

quick way to enable multiple compilation *without* reusing contexts (still about 10x faster than separate startup)