feat: add queue metric processing_failed

[PDT42]

This introduces a new observable counter queue.processing_failed. This metric is collected per instance, per tenant, per service and tracks the total number of times CAP failed at processing a scheduled task. This is different from cold_entries, which tracks the number of tasks that will no longer be retried.

The counter is incremented each time the attempts field of a given task is updated.

Considered Alternative Approaches:

• Handling the {event}/#failed event

  ... is not an option since it will only be emitted once processing failed finally

• Handling req.on('failed')

  is not an option, because the initial request on the queued service does not fail

• Calculating processing_failed based on incoming_messages and outgoing_messages

  ... is not an option because we allow to set attempts to maxAttempts manually for unrecoverable failures

• Calculating AND handling {event}/#failed

  ... could be an option, but registering the handler for each individual event feels super messy