Subgraph recover from non-determinstic error doesn't update status

[evaporei]

Definition

Subgraphs that fail for non-deterministic reasons, whenever they recover, their status doesn't get updated to failed: false.

Example

Let's say subgraph X failed for non-deterministic reasons at block 120. Everytime graph-node restarts, we'll process blocks up until 120 until it works. Since we're talking about a non-determinstic error, it might work on a second/third/... time.

The bug is that whenever that recover happens, the subgraph continues marked as failed, even though it's deployment block pointer is already at a number > 120.

How to fix it?

Our unfail function now only works for determinstic errors, it should consider those (non-deterministic) as well.

On a second note, @leoyvens suggested a retry strategy that doesn't require graph-node's restart. It isn't required for the bug to be fixed, but it's a great improvement. This will be tracked here: #2945

Related to: #2885