The buffer sizes in opaque case were computed by a mixture of hardcoded values
and mmal defaults. This caused breakage due to changes of the default values
in recent mmal versions. To avoid this in the future, the buffer sizes are
completely hardcoded now. As we have a shared header for mmal now, move the
required define for shared values into that header.
Using this new computation we can reduce the buffer size for opaque case by 2,
which saves some memory.