Physical link statistics should be much more densely sampled

[bnaecker]

This discussion came up with @rmustacc while developing OxQL. We both agreed that the current sampling interval of 10s for link statistics is effectively useless, and we should probably be much more aggressive here. At most 1s seems appropriate, but we may want to go even lower, though I need to do some quick experimentation to see how expensive that for the sled-agent and in terms of data storage in ClickHouse.

[karencfv]

Would it be possible to perform the same experiments on a replicated set up as well? I'd be interested to see how it performs compared to a single node.