Fix decode-only: return TTA suffix so pipeline runs decoding on loaded predictions

saved_prediction_path was returning a non-TTA suffix, causing the pipeline
to treat it as a final prediction and skip decoding entirely.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: Donglai Wei

connectomics/training/lightning/model.py

@@ -604,7 +604,9 @@ def _load_cached_predictions(
                 pred = read_volume(str(pred_file), dataset="main")
                 if pred.ndim < 4:
                     pred = pred[np.newaxis, ...]
-                return pred, True, cache_suffix
+                # Return a TTA-like suffix so the pipeline treats this as
+                # intermediate predictions (runs decoding), not final.
+                return pred, True, "_tta_x1_prediction.h5"
             else:
                 raise FileNotFoundError(
                     f"inference.saved_prediction_path not found: {pred_file}"