Add Horizon metric for keeping track of slow ingestion restarts

[tamirms]

While rolling out the limited history horizon instance to production we realized that some of the production ingesting horizon nodes were configured to run captive core with BucketDB disabled. Consequently, captive core was running with in-memory mode and this resulted in Horizon being very slow to resume ingestion after restarts.

We should have a metric which is incremented whenever captive-core has to catch up from scratch instead of quickly resuming from the LCL recorded in BucketDB. Once this metric is in place, we can modify our release testing checklist to make sure that the release branch of horizon does not regress by restarting captive-core from scratch at an unusually high rate.