Stopping the peer does not interrupt retry loops

[sundbry]

This plugin can hang in situations where the write-batch is stuck in a long loop, especially if you have a long retry timeout period. When the instance is (stop)ped, however, it doesn't interrupt this loop, and it can stick around still running in the background. This has some negative consequences

1. It prevents jobs from checkpointing their output state and you notice the error when you try to resume
2. it causes issues with background threads when reloading a lot during development

We could mitigate this by setting a smaller retry timeout, but I like to have a long one (especially in production) to cover any down time windows in external services to give them time to recover.