I don't know if 64 kB headers is such a good idea. You can put any metadata into message payload today, with incredibly
high message payload limits (512 MiB in future RabbitMQ versions, 2 GiB today), with any protocol.
You can have competing consumers on a queue today, with any protocol. My guess is that with shared subscriptions
each subscriber can consume a copy of the message.
The earliest release for which MQTTv5 would be considered is 3.9 (no ETA promises), in part because features such as
shared subscription would require significant changes to certain internal APIs first since that's not how the protocols we support
today work.