Some decoders can only handle specific endianness or a fixed depth and this allows better negotiation. Fixes bug #644208.
Still some issues, e.g. with seekable queries in totem, but also processing already-chunked input (created with matroskademux ! gdppay).