Failed flag should be cleared even if there is no fatal error

[leoyvens]

We've seen a few times subgraphs that are syncing fine, but have the failed flag set and no fatal error. I'm not sure how this happens, it probably shouldn't, but one thing we can easily address is that the unfail logic is not clearing the failed flag in this situation.

Here it is assuming that if there is no fatal error then the subgraph is not failed:

graph-node/store/postgres/src/deployment_store.rs

Lines 1437 to 1442 in ee76a63

	// We'll only unfail subgraphs that had fatal errors
	let subgraph_error = match ErrorDetail::fatal(conn, deployment_id)? {
	Some(fatal_error) => fatal_error,
	// If the subgraph is not failed then there is nothing to do.
	None => return Ok(UnfailOutcome::Noop),
	};

[leoyvens]

What gets a subgraph into this situation could be a rewind or revert, which will delete errors but not set a failed subgraph back to healthy. See #3978.

We should address that root cause first and see if it stops this from reproducing.