[server] Improve maintenance of workspace instance state

[svenefftinge]

When there's an exception happening in actuallyStartWorkspace we catch it and treat the start request as failed by updating the workspace_instance in the DB and marking it as stopped. The request to ws-manager might have succeeded, and the error might have been thrown later (the concrete case was a protocol error with billingservice).

We should try to keep the system consistent and try to cancel the workspace start with ws-manager in failInstanceStart.

Because the workspace instance was actually started, the stopped state will eventually be overwritten by ws-manager-bridge. It however only sets the status but doesn't reset the stoppedTimeand stoppingTimefields. If they are set, we should log an error but also set those two properties to their default state again.

[kylos101]

Thank you @svenefftinge ! As a heads up, there is a similar state management issue that can cause workspaces to get stuck in Pending, too. Might be a good use case to test after this change is made.

cc: @laushinka