If k of n authorities are down, k/n bootstrapping clients are delayed for minutes

When an authority is down, a client bootstrapping into the network might pick it to learn the consensus, and then have to wait until that attempt times out.

We should try to improve the robustness of the authorities. Fine, but easier said than done.

At the same time, we should explore letting clients time out more quickly during critical-path directory requests.

We should make sure not to screw it up for clients that are on genuinely slow and high-latency connections. The goal is to improve time-to-bootstrap for the average case while not harming it (much) for the already bad cases.