Upper bound on metrics batcher

[alexander-alderman-webb]

We currently have a threshold at which we start flushing metrics. Crossing the threshold does not prevent adding new metrics, leading to a memory issues when we add metrics at a faster rate than we can flush them.

Add a second, higher threshold at which we discard any incoming metrics. Also, log a message when we drop a metric because of the batcher size limit.

[linear]

PY-1958