Add _maybe_clear_multiplexer method to deal with memory leaks (#2642)

Summary:
Pull Request resolved: #2642

Adds _maybe_clear_multiplexer method, which is a no-op on base TensorboardMetric, to be implemented on its children where more is known about the multiplexer used.

Reviewed By: mpolson64

Differential Revision: D60935886

ax/metrics/tensorboard.py

@@ -159,6 +159,8 @@ def bulk_fetch_trial_data(
                                 "the curve empty in the TensorBoard UI?"
                             )
 
+                    self._maybe_clear_multiplexer(mul)
+
                     df = (
                         pd.DataFrame(records)
                         # If a metric has multiple records for the same arm, metric, and
@@ -222,6 +224,15 @@ def _get_event_multiplexer_for_trial(
 
             return mul
 
+        def _maybe_clear_multiplexer(
+            self, mul: event_multiplexer.EventMultiplexer
+        ) -> None:
+            """Clear the multiplexer to avoid memory leaks. This is a no-op on the base
+            class, to be implemented on child classes where more is known about the
+            multiplexer.
+            """
+            pass
+
 except ImportError:
     logger.warning(
         "tensorboard package not found. If you would like to use "