Add a Prometheus metric for measuring the scalable object loop processing deviation

[JorTurFer]

Proposal

@MrsDaehin and I are working on #4411 but we have noticed that there isn't any metric
(in the operator) that we can use for measuring how KEDA is performing in terms of response time.

As the majority of the work (during the normal operation) is done here:

keda/pkg/scaling/scale_handler.go

Lines 164 to 181 in bce3375

	for {
	tmr := time.NewTimer(pollingInterval)
	h.checkScalers(ctx, scalableObject, scalingMutex)
	select {
	case <-tmr.C:
	tmr.Stop()
	case <-ctx.Done():
	logger.V(1).Info("Context canceled")
	err := h.ClearScalersCache(ctx, scalableObject)
	if err != nil {
	logger.Error(err, "error clearing scalers cache")
	}
	tmr.Stop()
	return
	}
	}
	}

I think that we can add a metric with the difference between the calculated time for next execution (now + pollingInterval) and the real execution time. If the time is close to 0 it means that we are fine, but if the new loop starting is too much later than the expected, we aren't executing the control loop on the expected time, which probably is because the CPU is overloaded

Use-Case

This metric can give as a measure about how the load is impacting in the scalable object processing loop measuing the time between loops

Is this a feature you are interested in implementing yourself?

Yes

Anything else?

No response