Our current implementation sends an individual message for each advertised prefix (NLRI). This is inefficient when we have bursts of routes coming in (such as new peer coming up), where it is likely that prefixes have the same set of attributes.

Investigate how this packing could be achieved and where: it we needs to understand how big the message can be (e.g. 4K or 64K) and it also needs to know how big the attributes will look on the wire and how adding more NLRIs is going to affect the message. BGPMessageToByteEncoder may be the right place, but it is currently stateless and has little understanding of the message structure.