Remove dead recoverable_exceptions and is_recoverable_fetch_e (facebook#5120)

Summary:
Pull Request resolved: facebook#5120

Follow-up to D98924467, which decoupled metric fetch errors from trial
status in the Orchestrator. The orchestrator no longer uses
`recoverable_exceptions` or `is_recoverable_fetch_e` to decide trial
fate, making them dead code.

- Remove `Metric.recoverable_exceptions` class attribute and
  `Metric.is_recoverable_fetch_e` classmethod from `ax/core/metric.py`.

Reviewed By: bernardbeckerman

Differential Revision: D98932195

Author: saitcakmak

ax/core/metric.py

@@ -88,11 +88,6 @@ class Metric(SortableBase, SerializationMixin):
         properties: Properties specific to a particular metric.
     """
 
-    # The set of exception types stored in a ``MetchFetchE.exception`` that are
-    # recoverable ``orchestrator._fetch_and_process_trials_data_results()``.
-    # Exception may be a subclass of any of these types.  If you want your metric
-    # to never fail the trial, set this to ``{Exception}`` in your metric subclass.
-    recoverable_exceptions: set[type[Exception]] = set()
     has_map_data: bool = False
 
     def __init__(
@@ -164,17 +159,6 @@ def period_of_new_data_after_trial_completion(cls) -> timedelta:
         """
         return timedelta(0)
 
-    @classmethod
-    def is_recoverable_fetch_e(cls, metric_fetch_e: MetricFetchE) -> bool:
-        """Checks whether the given MetricFetchE is recoverable for this metric class
-        in ``orchestrator._fetch_and_process_trials_data_results``.
-        """
-        if metric_fetch_e.exception is None:
-            return False
-        return any(
-            isinstance(metric_fetch_e.exception, e) for e in cls.recoverable_exceptions
-        )
-
     # NOTE: This is rarely overridden –– oonly if you want to fetch data in groups
     # consisting of multiple different metric classes, for data to be fetched together.
     # This makes sense only if `fetch_trial data_multi` or `fetch_experiment_data_multi`