Attached patch modifies unicode escape and raw unicode escape encoders to use the new _PyBytesWriter API.

The patch is optimized to encode Latin1 characters: encoding Latin1 characters when no character is escaped should not have to call _PyByte_Resize() at all.

When characters are escaped or a BMP or non-BMP string is encoded, overallocation is used to reduce the number of _PyByte_Resize(). It uses _PyBytesWriter overallocation strategy instead of always overallocate for the worst case.

_PyBytesWriter also embeds a small buffer allocated on the stack which also avoids calls to _PyBytes_Resize() when the output fits into 512 bytes.