fix: reset the epoch-dedup tracker per run

Reset `_logged_epoch_step` in `reset_metrics` (called once per fit/validate/test
run, not per epoch) so the convenience `epoch` metric is re-emitted at the start
of a fresh run instead of being suppressed when a standalone validate/test happens
to reuse the same (step, epoch) as the preceding fit.

Author: gaurav0107

src/lightning/pytorch/trainer/connectors/logger_connector/logger_connector.py

@@ -241,6 +241,9 @@ def reset_metrics(self) -> None:
         self._progress_bar_metrics = {}
         self._logged_metrics = {}
         self._callback_metrics = {}
+        # reset per-run so the convenience `epoch` metric is re-emitted at the start of a
+        # fresh fit/validate/test run (this is called per-run, not per-epoch)
+        self._logged_epoch_step = None
 
     def reset_results(self) -> None:
         results = self.trainer._results