Action refinements

[ageorgou]

Some fine-tuning for how actions are performed to make their display and handling smoother:

• [x] Don't auto-update the status of hosts being destroyed.
• [x] If a host's status is Error, deleting it should delete the whole resource group ("hard delete"), because deployment may have stopped at the VM stage.
• [x] If a host's status is NotFound, deleting it should not call terraform destroy. We should instead check whether the group exists and, only if it still does, delete the whole resource group. This avoids the problem of a user trying to delete a machine which has already been deleted (e.g. through the Azure Portal).
• [ ] We should check whether the chosen actions are applicable (e.g. cannot restart a stopped machine). This is important because users could potentially request an action by giving the corresponding URL, even if the dashboard button is disabled.

More to be added if found.

[ageorgou]

Most items covered in #80. We can revisit the URL checking if it becomes an issue.