dcp_proxy should optimize the quantity of copying done when dealing with large mutations

DCP proxy doesn't manage DCP mutations it receives optimally. In particular, large mutations are read in chunks and are appended in sequence to the mutation binary so that the total DCP mutation "packet" can be passed as a convenient blob to routines that process it. These routines simply write the packet to the peer proxy socket in most case, but some mutations do indeed need to be completely decoded. Of course, these mutations are always the small DCP control mutations.

Optimizing how this is handled can result in lower CPU and lower memory use that will be most noticeable in situations with large mutations.