TCP_NODELAY / Disable Nagle's algorithm

[decarbonise]

amqprs looks really good and the API seems nice to work with. On the performance side, Lapin exposes the ability to enable TCP_NODELAY / disable Nagle's algorithm by configuring the socket directly before connecting. I can't see the ability to do so in amqprs, but if there is a way to add it it could substantially reduce latency for small, frequent, latency-sensitive messages.

[michaelklishin]

We should make sure that the Nagle's algorithm is disabled without configurability.

In my close to 17 years of RabbitMQ contributions, I am yet to see a workflow that would benefit from Nagle's algorithm.

[michaelklishin]

Voilà #184.

[decarbonise]

Such a speedy PR! Completely agree re: having Nagle always disabled.