Previously, the sidekiq metrics middleware would not record latency
and cpu duration metrics for failed sidekiq jobs.

This change does two things:
1. Ensures that Sidekiq jobs that fail are also included in latency
   histograms.
1. Adds a `job_status` label to Sidekiq latency metrics so that failed
   job latencies can be differentiated from successful jobs.