fix: Add noop to request errors during cleanup

[afharo]

We noticed a potential race condition that's hard to reproduce (it only happened 29 times in the last 30 days in our production environment). Kibana crashes with EPIPE during write operations (task manager writing the result of a task, task manager claiming tasks, Kibana during the setup phase, and multiple POST requests).

The lack of stack traces in Kibana (elastic/elasticsearch-js#2235) doesn't help to identify if it's related to one specific request or spread across multiple ones. But the gut feeling is a race condition where potentially, multiple errors are emitted at once and only one is caught, cleaning the listener, and the next one crashes the app.

Happy to close this PR if you think that this change is not safe/sensible.

[JoshMock]

The effects of this change are potentially messy, mostly because it's tricky to map all the possible paths through the event listeners in HttpConnection. I'm just back from PTO so I want to come back to this and see if some new unit tests, and manually testing this change against the client's integration test suite, would increase confidence.