Improve catchup

[jbearer]

Anticipating some problems we might run into as more nodes start pointing at more peers for catchup:

• HTTP timeouts can be really bad because it can make us take a really long time (ie more than a view) to even get to all the peers in our list
• There is not currently any intelligent way of ordering peers, so if you have a really unreliable peer early in the list, that can cause performance issues

This PR:

• Implements an aggressive and adaptive timeout mechanism so no one faulty peer can cause us to timeout before trying to catch up from other peers
• Ranks peers based on request failure rate so we always try our most reliable peers first
• Adds metrics for catchup requests