Mesure the performance impact of precalculated counters

In order to chose a solution for #373 (closed) we need to know if precalculating most counters makes a real performance difference.

My intuition would be that it makes the insertion a bit more costly (maybe even quite a bit given the number of triggers), but this is amortized once the initial scan is complete.

AFAICS we need to:

• Select a trigger to evaluate (in order to avoid writing the equivalent requests for all counters)
• Mesure the impact at insertion
• Mesure the impact when reading