prevent redundant coercion from Py_UNICODE to a unicode string when the subscript index is a Python object - this case is no longer optimised

fix test py_unicode_type, remove code from Optimize.py that became redundant when Py_UNICODE started coercing to unicode by default

parser support for generator expressions so that we can give a better error message than "Expected ')'"

fix ticket #145 also for CondExprNode, make "bint (+) non-bint -> object" a general rule for independent spanning type finding, merge duplicated type determination functionality from CondExprNode back into PyrexTypes.py