Emit metric for failure to step jobs

It's possible for Janus to fail to step a collection job because it notices that there are pending aggregation jobs, or that there are relevant reports that have not yet been scheduled into a job. When this happens, we increment step_attempts and try again later. But then we have no indication of why we released the job, except that if it was due to failure to make a request to the helper or a DB problem, we'll log an error.

We should emit a metric from Janus that increments when we release a job and increment the step attempts counter. We could then label with a reason enum (pending_agg_jobs | unaggregated_report | http_request_falure| etc.). It can't be labeled by task or job ID to avoid high cardinality metrics, but at least then we'd notice a correlation between long collection job completion times and an uptick in that graph.

Start by locating the Janus code path that releases a collection job and increments step_attempts. Trace the existing reasons for release, including pending aggregation jobs, unaggregated reports, helper HTTP failures, and database problems. Done means a metric is emitted for these failures with bounded reason labels and without task or job IDs.