When a control plane property read happens and we are reading the status register to
test whether there were worker-level errors, a timeout value from the worker does not set the sticky timeout-seen bit in the worker status register.
Secondarily, there is no global way to force these status-register-reads to happen after all
accesses (which slows things down). There should be a global or container-level or worker-level setting which does the extra error check all the time for convenient debug.