Introduce a new RTS call for OS.IO.poll, and incidentally OS.Process.sleep, that waits for only a short
period. Retrying for longer periods is done in ML code. Retrying in the RTS itself has been a source of problems
in the past.