This allows concurrent decoding of the last field/frame, rather than only the last slice, of data packets with multiple NAL units packed together. This will fix the slowdown reported in e.g. bug 52. Signed-off-by: Anton Khirnov <anton@khirnov.net>tags/n0.9