Implement aligned sampling for ZMON checks

[mohabusama]

Random sampling implemented in #382 is aimed at throttling workers rather than more accurate aligned datapoints that could be used later in aggregation function (min(1m), avg(1m), sum(1m), etc...).

The problem with random sampling is that sampled check-entity results are not aligned and could lead to fluctuating/misleading graphs/metrics.

A possible solution would be deciding on sampling while taking into consideration check_id, entity_id, check interval.

fn(check_id, entity_id, interval, alert_state, alert_changed)