vm-runner: Avoid alerting when op-node restarts

Update vm-runner and our alert monitoring to ensure that op-node can restart without triggering any alerts. There are two parts to this, both will probably need a bit of thinking about to work out what the actual right answer is:
1. Don't fire setup failure alerts so eagerly. We really only care about setup failures if they are happening persistently (vs invalid status codes which should always alert). Should still alert if one trace type is consistently failing to setup and others are working.
2. Just after op-node starts up it has pretty unreliable sync status output. The finalized L1 hash is 0x0000... and the CurrentL1 may be below what is required to actually derive the reported current L2 safe head. Possibly need to just wait until this resolves itself or may be able to take the CurrentL1 but then query the safe head db to find the L2 block to use instead of trusting the sync status.

Start by tracing vm-runner's setup-failure alerting and op-node sync-status handling. Reproduce an op-node restart if possible, then define and verify behavior for persistent setup failures, invalid status codes, and startup values that are not yet usable; done means restarts no longer trigger false alerts while genuine failures still do.

Written by the indexing model from the issue text.