More editing is probably desirable.

If maxchar == PyUnicode_MAX_CHAR_VALUE(unicode), we do an useless copy.

 * If needle is 0, try (p[0] >> 16) & 0xff for UCS4
 * Disable fastsearch_memchr_1char() if needle is zero for UCS2 and UCS4

Truncate the buffer size to PY_SSIZE_T_MAX.

Use Py_off_t type (64 bits) instead of off_t (32 bits).

Instead, the re module's standard caching does its work.

(thanks "make smelly")