Resolve flakyness in test_kill_before_submit

Sometimes the job actually starts on a compute node and sometimes
it does not before it is removed/killed. In the the latter case, code
previous to this commit will fail as it asserts that the job script
has actually started to run.

tests/ert/unit_tests/scheduler/test_lsf_driver.py

@@ -1326,7 +1326,10 @@ async def finished(iens: int, returncode: int):
         # detail we do not want to track.
         assert returncode in (SIGTERM, SIGNAL_OFFSET + SIGTERM, LSF_FAILED_JOB)
 
+        if returncode != LSF_FAILED_JOB:
+            # We will only see the was_killed file if a compute node
+            # got a chance to start to job script:
+            wait_until((tmp_path / "was_killed").exists, timeout=4)
+
     await poll(driver, {0}, finished=finished)
     assert "ERROR" not in str(caplog.text)
-
-    wait_until((tmp_path / "was_killed").exists, timeout=10)