Update deployment_failed to better reflect syncing status

[azf20]

The deployment_failed metric is only created/updated once the block stream has been set up, but not when starting up a deployment fails, e.g. because it references a non-existent chain

This makes it harder to detect failed subgraphs. We should update (and maybe rename) this metric so that it better captures syncing status

starting - a node figured that it should run the deployment and is working on starting it; that should be set as early as possible
running - everything's fine, we are processing blocks
failed - something went wrong