[FEATURE] Discuss feature logging architecture in documentation

[JohannesDaniel]

The current documentation suggests to combine user queries with feature logging, which has been repeatedly reported to increase query latencies significantly (see also #50).

This trap can be avoided by discussing an alternative approach considering feature logging rather as a batch job running once a day instead of permanently running besides querying and interfering regular user traffic. In several cases, this will lead to significantly reduced latency, but also to less precise logging data.

The documentation should be enhanced by discussing this tradeoff, also considering typical short-head/long-tail distributions.

[sandeshkr419]

@JohannesDaniel Do you plan to get this documentation improvement change out?

[JohannesDaniel]

@sandeshkr419 yes, will do as soon as I find the time :)