#943

My attempt at implementing automatic reprepare+retry of prepared statements that hit a cache error.

I thought the place place to do this would be in the server impl. That way, we can avoid dealing with sharding and duplicating the retry logic at different points. Please let me know if I should move it elsewhere though.

The main piece of work here is a small state machine in each server that keeps track of what the current batch of messages are, and if they've returned with expected. That way, we can know which messages we need to retry, and which messages we shouldn't send again. An example being

Parse
Describe
--- later ---
Describe
Bind <-- cache error happens here
Execute
Sync

In this case, we want to reprepare and only retry from bind onwards.