Prevent unnecessary remove/re-add of bad nodes

[zacstewart]

Instead of ejecting bad nodes and potentially having them re-added later, we should just stop using them, and if they become good again, use them then.

The problem is that slow nodes may fail to respond to queries and thus become bad and get rejected. They may eventually respond, but by that time, we're not expecting their response so we discard it.

It also produces a lot of volatility in the network with constant re-introduction of nodes.

This will make updating the socket addr of dead nodes important