Increase operator resiliency to unexpected scaler failures about keda

[JorTurFer]

This is an interesting point. We aim to prevent all the panics by code instead of just recovering them, but maybe we could recover panics on scaler metric requests:

keda/pkg/scaling/cache/scalers_cache.go

Lines 125 to 142 in 1e1cfb1

	func (c *ScalersCache) GetMetricsAndActivityForScaler(ctx context.Context, index int, metricName string) ([]external_metrics.ExternalMetricValue, bool, int64, error) {
	if index < 0 || index >= len(c.Scalers) {
	return nil, false, -1, fmt.Errorf("scaler with id %d not found. Len = %d", index, len(c.Scalers))
	}
	startTime := time.Now()
	metric, activity, err := c.Scalers[index].Scaler.GetMetricsAndActivity(ctx, metricName)
	if err == nil {
	return metric, activity, time.Since(startTime).Milliseconds(), nil
	}
	ns, err := c.refreshScaler(ctx, index)
	if err != nil {
	return nil, false, -1, err
	}
	startTime = time.Now()
	metric, activity, err = ns.GetMetricsAndActivity(ctx, metricName)
	return metric, activity, time.Since(startTime).Milliseconds(), err
	}

As scalers are the place where more contributions are made, they're also the place with more unexpected problems and although I think that we should avoid panics, maybe in this case it can make sense.

In the other hand, we have a really few panics because we try to cover all the cases and we almost achieve it.
WDYT @zroubalik @dttung2905 ?