The new API allows (optionally and on by default) using a internal buffer to encode, avoiding
the need to allocate large buffers or risking failure on too small buffers.
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>