Cluster should always aggregate metrics in a same order

[matej21]

The issue with the current metrics aggregation is that it occurs in the order in which individual workers respond to the GET_METRICS_REQ message. This may cause variations in the aggregation results when the metrics contain floats, such as a histogram.

For example, if workers respond to the first request with values 0.5848208, 0.5479198, 0.3437699 (which sum to 1.4765105), and then respond to a second request with the same values but in a different order (0.3437699+0.5848208+0.5479198), the result of the aggregation will be1.4765104999999998 in JS, due to float errors.

This can trigger a "reset" detection in Prometheus and severely impact the accuracy of the graphs.