Do not use average when calculating total throughput/latency of backups

Currently we use a simple average when calculating the total throughput/latency of backupo instances to compare it with the master ones.

This may lead to problems of false positive View Changes when there is a spike on just one of a backups.

Acceptance criteria

• Define a better way to calculate the average for backups. Consider median which is more robust to spikes.
• Write unit tests checking, in particular, that spikes should not trigger false positives.
• Implement the changes