Added an explicit cast in c_code for AdvancedSubtensor1 Op to match logic in perform() and avoid test failures
on 32 bit platforms.  The real issue seem to be that x.shape is a 64-bit tensor regardless of the size of platform's
 size_t type.  This leads to int64 type appearance in any computation involving shape and resulting index cannot
 be passed back to numpy.take() necessitating the cast.