Drop Adaptive._retire_workers

The `_retire_workers` method seems to largely duplicate the same in
`distributed`'s `Adaptive`. So just go ahead and drop our
implementation.

This was tried before, but didn't work do to duplicate `retire_workers`
calls to the `Scheduler` in both `Adaptive._retire_workers` and
`DRMAACluster.scale_down`. However as the behavior of
`DRMAACluster.scale_down` has now been corrected, it should now be
possible to drop our implementation of `Adaptive._retire_workers`. Hence
we now do drop it here.

dask_drmaa/adaptive.py

@@ -94,20 +94,3 @@ def get_scale_up_kwargs(self):
         logger.info("Starting workers due to resource constraints: %s",
                     kwargs['n'])
         return kwargs
-
-    @gen.coroutine
-    def _retire_workers(self, workers=None):
-        if workers is None:
-            workers = self.workers_to_close()
-        if not workers:
-            raise gen.Return(workers)
-        with log_errors():
-            result = yield self.scheduler.retire_workers(workers,
-                                                         remove=True,
-                                                         close_workers=True)
-            if result:
-                logger.info("Retiring workers {}".format(result))
-            # Diverges from distributed.Adaptive here:
-            # ref c51a15a35a8a64c21c1182bfd9209cb6b7d95380
-            # TODO: can this be reconciled back to base class implementation?
-        raise gen.Return(result)