Move the 'buffered' reading into shared-memory-ring

[djs55]

To be able to survive crashes we need to read whole packets into a persistent buffer, which is tricky given that the ring is too small to accommodate a whole packet.

To achieve this we seem to have created a not-very-good circular buffer implementation which is supposed to be layered on top of the quite-decent implementation in shared-memory-ring. Perhaps we should generalise shared-memory-ring to be able to create larger ring buffers.