Make task-killing tests more reliable

test_execute_cmd_finished_before_killpg_sigkill assumed that its test subprocess would finish exiting between the SIGKILL sent by the test (due to wait_popen_sigkill=True) and the SIGTERM sent by RunCommandTask._run_cmd. This was typically the case, but not always; quite small perturbations such as adjusting the short timeout used by RunCommandTaskForTesting._wait_popen could disturb it, and it sometimes failed in CI.

The failure was just because RunCommandTask._run_cmd didn't log "aborted:" and "returncode:" in the case where the task was aborted and the subprocess didn't finish before _run_cmd sent SIGTERM to it. Try harder to force this scenario in tests, and fix the logging.

Fixes: #458 (closed)

Closes: https://bugs.debian.org/1077903