system_command: fix potential issue of stderr not being read

[Bo98]

This is a blind, possible fix for test flakiness surrounding stderr in the system_command tests. I was not able to reproduce the issue locally but did identify a possible cause.

The potential bad flow I've identified here:

1. Assume ProcessTerminatedInterrupt is received before IO.select signals readable streams (very narrow window that I cannot reproduce - maybe will happen in VMs)
2. On receiving this, readable_sources is set to all streams and pending_interrupt is set to true
3. We run readable_sources.none? to loop through and read all streams. However none? as it was written was guaranteed to short-circuit after the first stream as all possible paths would lead it to returning true (EOF on first stream will return value of sources.any? which will be true given we haven't got to the second stream yet).
4. We exit the until loop because of pending_interrupt with that second stream not being read yet.

The correct behaviour should be to read all readable_sources so I've adjusted the logic slightly to make sure that is the case.

[MikeMcQuaid]

Worth a try, thanks @Bo98!