entity_addr_t comparison uses memcmp improperly (endianness bug across the wire)
We use a memcmp on the entity_addr_t object for its comparator, and we use that in resolving connection races within the messenger.
However, because memcmp does a byte-by-byte comparison, this can return different results if run on big-endian and little-endian architectures! That would be bad. It needs to be fixed reliably, but that will require a protocol revision or feature bit or something to make sure it's safe. (Hat-tip to Sam for noticing this by inspection while searching for another bug!)