spire virt auto launch should not scan for supervisor keys

[celskeggs]

The code used in spire virt auto install to make sure we have the right keys for the supervisor node is not needed -- and, in fact, is actively unwanted -- during launches of existing clusters.

We should change spire so that this code is only used when it's needed.

[cryslith]

I don't quite understand - why does the ssh command succeed during install but fail during a subsequent launch?

[celskeggs]

I'm not entirely sure off the top of my head why it fails, but the reason we shouldn't need this is that, after the first launch, we should have everything we need set up to authenticate correctly without checking the hostkey via the MOTD fingerprint.

[cryslith]

Additional context from meeting: it subsequently fails because we intentionally shift aside old authorized_keys after we set up the supervisor's final ssh configuration. So that part is intended behavior.