Run shell commands reliably

Tests sometimes fail due to a shell command failing, bet we don't get the
appropriate result from the command run itself, which leads to confusing
states and challenging troubleshooting.
E.g. jobscheduler reports its failure only via stderr.

This introduces a utility that fails with the relevant command output,
should something be printed to stderr.

Test: atest AutoRevokeTest
Bug: 168843991
Change-Id: I5e354c1b88ca3ceeeda2d5500b62c421174fd57c