Relax the checks/assertions on which PID owns the semaphore (#822)

It turns out that the checks to ensure which PID owns the semaphore
is too aggressive. The PID might be 0, pid of our own process or
any process above to our PID in the process tree.

So, we simplify the rule. We only terminate the semaphore when
we are owner of the PID. Else, we wait until the owner of the
PID exits (which will eventually happen).

Author: onderkalaci

src/bin/pg_autoctl/lock_utils.c

@@ -107,27 +107,14 @@ semaphore_finish(Semaphore *semaphore)
 	 * its semId as found in our environment variable PG_AUTOCTL_LOG_SEMAPHORE.
 	 *
 	 * At finish time (called from the atexit(3) registry), we remove the
-	 * semaphore only when we are the owner of it. We expect semaphore->owner
-	 * to be either zero (0), or to have been filled with our own pid.
+	 * semaphore only when we are the owner of it.
 	 */
-	if (semaphore->owner == 0)
-	{
-		/* there's no semaphore closing protocol in SysV */
-		return true;
-	}
-	else if (semaphore->owner == getpid())
+	if (semaphore->owner == getpid())
 	{
 		return semaphore_unlink(semaphore);
 	}
-	else
-	{
-		log_fatal("BUG: semaphore_finish semId %d owner is %d, getpid is %d",
-				  semaphore->semId,
-				  semaphore->owner,
-				  getpid());
 
-		return false;
-	}
+	return true;
 }