Example case on how this can happen:

1. The ARS yields
2. ARS waits for response and got an error (for example, stale config error)
3. ARS tries to unyield and gets an error while unyielding.
4. The unyield error gets pushed to the response queue for the rest of the remotes.
5. However, since the stale config exception was pushed to the response queue first, it gets popped first.

We should not be treating unyield errors as remote errors since it came from the local process and we should instead make sure to bubble up the yield error when it happens.