[aggregator|tcpclient] Drop enqueued payloads for a flush after write fails

[vdarulis]

…failed

What this PR does / why we need it:

If an instance is slow/dead and writes (with retries) fail, don't just abort flush - drop all enqueued payloads (captured at the time of Flush()). Otherwise, the backlog is accumulating up to the point the queue is dropping upstream and causes even more load on the target aggregator instance the next flush cycle.

Special notes for your reviewer:

Does this PR introduce a user-facing and/or backwards incompatible change?:

Does this PR require updating code package or user-facing documentation?: