Some question have come up regarding what metrics we collect on the router for query stats when it comes to writes.

What is "execution time" from the router's perspective, considering that we do batching on the router? Ditto for some of the other time-related metrics.
When should we fold in metrics we collect for the actual single operation on the router? When does it make sense to sum of metrics from the shards vs. doing something else (taking a max for example)

We should answer these questions such that we follow the principle of least surprise, but also provide useful metrics, and then implement that.