The added test, intended to fail and reproduce the [reported
issue](facebook#28595), unexpectedly
passes in its current state. I see three possible reasons:

1. The bug report could be invalid.
2. How I've structured the test might be insufficient to replicate what
   `ai/rsc` is doing.
3. Something in the test setup could be masking the actual error.
   (Maybe related to fake timers?)

If the problem lies in reason 2 or 3, this test could possibly serve as
a foundation for further investigation.

fixes facebook#28595