Set up an aggregation buffer ring and configure placement mode so data
that doesn't fit in the rx buffer goes into an aggregation buffer,
allowing jumbo frames to be received. Using 8k aggregation buffers
means we'll only ever need one per packet.
When receiving jumbos, sometimes we get an interrupt before all three
of the completion events are ready, in which case we should not consume
the events that are ready. Expanding the completion ring makes this
happen less frequently, so allocate four cp ring pages instead of one.