There are many places where I could add a mutex deadlock and the tests wouldn't catch it, which seems likely to happen in production scenarios.

This code should be almost 100% covered since a deadlock kills all future workspace connections until a restart. We should decompose this code into a more testable fashion and maybe change the structure of the ServeXXX for easier testing.

Cases to cover:

• >80% coverage
• Database disconnect with HA coordinator for pubsub
• Client connects before agent on different servers
• Agent connects before client on different servers
• Block massive node payload from being sent
• Rate-limit nodes being sent from either end (maybe debounce at a low latency instead?)
• Ensure AsOf is respected when entering new nodes from clients, agents, and pubsub