@pavel-kirienko
... if we support scattered buffers, DSDL serialization could be sped up considerably in cases where there are large byte arrays present in the message (like imagery data) because the serializer would be able to serialize only non-trivial parts of the message and keep references to the blobs:
If we go this way, to support full end-to-end low-copy operation, the application would need to allocate the message object itself from the media memory resource.
One side effect of scattered input buffers is that an output frame may end up straddling multiple fragments; to support this we will need to allow a tx queue item to reference not just one but an arbitrary number of fragments from the input buffer.
See https://github.com/OpenCyphal-Garage/libcyphal/pull/343