Training a curation model with metrics from older version fail

I am trying to train a model for curation with an analyser computed with version 0.103.0, using spike interface version 0.104.8 fails with error

KeyError                                  Traceback (most recent call last)
Cell In[22], line 3
      1 folder = "/Users/layardh/The Francis Crick Dropbox/Hugo LayardHorsfall/1 SUNAN/OSSS_unitrefine/UnitRefine/human_np"
      2 
----> 3 trainer = sc.train_model(
      4     mode="analyzers",                           # Use SortingAnalyzer objects as input
      5     labels=[labels],                            # List of label arrays (one per analyzer)
      6     analyzers=[analyzer],                       # List of SortingAnalyzer objects

File ~/The Francis Crick Dropbox/Hugo LayardHorsfall/1 SUNAN/OSSS_unitrefine/UnitRefine/.venv/lib/python3.13/site-packages/spikeinterface/curation/train_manual_curation.py:762, in train_model(mode, labels, analyzers, metrics_paths, folder, metric_names, imputation_strategies, scaling_techniques, classifiers, test_size, overwrite, seed, search_kwargs, verbose, enforce_metric_params, **job_kwargs)
    760 if mode == "analyzers":
    761     assert analyzers is not None, "Analyzers must be provided as a list for mode 'analyzers'"
--> 762     trainer.load_and_preprocess_analyzers(analyzers, enforce_metric_params)
    764 elif mode == "csv":
    765     for metrics_path in metrics_paths:

File ~/The Francis Crick Dropbox/Hugo LayardHorsfall/1 SUNAN/OSSS_unitrefine/UnitRefine/.venv/lib/python3.13/site-packages/spikeinterface/curation/train_manual_curation.py:276, in CurationTrainer.load_and_preprocess_analyzers(self, analyzers, enforce_metric_params)
    273         metric_names = metric_extension.params["metric_names"]
    274         consistent_metrics = list(set(metric_names).difference(set(conflicting_metrics)))
    275         consistent_metric_params = {
--> 276             metric: metric_extension.params["metric_params"][metric] for metric in consistent_metrics
    277         }
    278         self.metrics_params[extension_name + "_params"]["metric_params"] = consistent_metric_params
    280 self.process_test_data_for_classification()

KeyError: 'mahalanobis'```

Start in spikeinterface/curation/train_manual_curation.py at train_model and CurationTrainer.load_and_preprocess_analyzers. Reproduce training with an analyzer computed by version 0.103.0 and SpikeInterface 0.104.8, then inspect how the mahalanobis metric is handled in metric_params. Done means this cross-version input trains successfully or reports a clear compatibility error.