Recognize early when an autocluster is started without an SSH agent

[celskeggs]

When the spire virt auto cluster command runs, it requires that an SSH agent exist for it to register SSH keys with -- but it doesn't check this until it's in the middle of configuring the supervisor node. If someone has forgotten to set one up, this means they need to re-run half of the process, which is very much a waste of time.

We should recognize the lack of an SSH agent immediately, possibly just by running ssh-add -l or recognizing whether the $SSH_AUTH_SOCK environment variable exists. This will save time.

[celskeggs]

Relatedly, there's also a problem where, if the SSH agent has too many already-loaded identities, it can fail to authenticate because it runs out of tries. It may be worth clearing out old ssh identities from the SSH agent automatically, which might be part of the same fix. (Or could be a separate fix. Either way is good.)

[cryslith]

@dihernandez is https://github.com/sipb/homeworld/tree/check_ssh_agent_early ready to PR?

[celskeggs]

@dihernandez @krawthekrow any progress on this?