[Bug]: Instances that are temporarily not available should be skipped during provisioning a job

[peterschmidt85]

Steps to reproduce

1. Add an instance, e.g. via dstack pool add-ssh, and wait until the instance is shown as idle
2. Shutdown the instance manually (via the cloud console or by plugging of the power)
3. Call dstack run with --reuse

Actual behaviour

1. The run is assigned to the instance (that is not available at the moment of submission of the run but still shown as idle)
2. The run gets stuck in provisioning

Expected behaviour

1. If the instance is temporarily unavailable, mark it with temproarily_unavailable=true (and show it in the console/UI)
2. Skip such instances during provisioning new jobs

dstack version

Any

Server logs

No response

Additional information

No response

[r4victor]

Especially relevant after #1200. Currently, dstack may resubmit the run to the cloud instance that is no longer available (e.g. was interrupted/deleted). The dstack will remove the instance only after it doesn't hear from the instance for 20m.

[r4victor]

@peterschmidt85, any preferences how display instances that are temporarily unavailable? We could use unavailable status but it can easily be mistaken with busy (both mean unavailable for provisioning). I suggest we show it as unreachable.