There has been a long standing issue where the exit code for detached processes was null - so we could never detect that they exited. This can cause builds to hang and continue to try scheduling work on a dead worker.

This fixes that by instead using the input stream (stdout of the worker) to detect a dead worker.