Improve Teleport availability and resiliency

[russjones]

At the moment we have the following problems with Teleport.

• there was in-memory cache that was only invalidated on server restart.
• in case if all auth servers are down, no new SSH sessions could be created

We should improve it in the following ways:

• teleport provides explicit control over cache TTL for Proxy and Node
• cache is persistent and on-disk
• in case if all auth servers are down, all systems including Proxy (reverse tunnels, etc) and Nodes are functioning
• introduces throttling for failed auth servers to avoid locks on connection timeouts
• introduces client-side caching for discovery data
• introduces new expires metadata property uniformly supported by all resource objects

[russjones]

Fixed in #853