Add metric to indicate that poll rate is limited by client

[mfateev]

Currently there is no metric which can indicate that poll rate is limited by client (for example maximum number of parallel tasks is reached), not by the service or connection latency.

There are multiple reasons for polls being slowed down. I'm not sure if we need a single metric with appropriate slowdown tag or multiple metrics. Here is the polling loop code.