Excessive number of stream resets

[notanatol]

Context

A random sample off of the testnet-storage cluster shows that in a span of 5 minutes there have been 698 log events with category level 'error' - of which 362 are in the form: "error"="SOME ACTION: stream reset"

Sometime the action is multiple errors joined, due to recent enhancements in error handling:

"error"="new stream: send headers: read message: stream reset"

Every protocol exhibits these stream resets and some seem to be due to accounting reasons.

Summary

This ticket keeps track of the efforts to investigate (and fix) the core reason(s) for these stream resets.