Fix confusing weight loading logging for legacy models

When loading backbone/head weights separately from a legacy .h5 file,
the logging showed confusing counts like "22/26 loaded" instead of
"22/22" because it used total legacy weights as the denominator.

Changes:
- Use PyTorch model parameters as denominator instead of total legacy weights
- Change "No matching PyTorch parameter" messages from warning to debug
  level since mismatches are expected when loading components separately
- Add debug message showing count of unmatched legacy weights

Before: "Successfully mapped 22/26 legacy weights to PyTorch parameters"
After:  "Successfully mapped 22/22 PyTorch parameters from legacy weights"

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

Author: gitttt-1234

sleap_nn/legacy_models.py

@@ -260,7 +260,7 @@ def map_legacy_to_pytorch_layers(
         if matching_pytorch_name:
             mapping[legacy_path] = matching_pytorch_name
         else:
-            logger.warning(f"No matching PyTorch parameter found for {legacy_path}")
+            logger.debug(f"No matching PyTorch parameter found for {legacy_path}")
 
     # Log mapping results
     if not mapping:
@@ -270,8 +270,13 @@ def map_legacy_to_pytorch_layers(
         )
     else:
         logger.info(
-            f"Successfully mapped {len(mapping)}/{len(legacy_weights)} legacy weights to PyTorch parameters"
+            f"Successfully mapped {len(mapping)}/{len(pytorch_params)} PyTorch parameters from legacy weights"
         )
+        unmatched_count = len(legacy_weights) - len(mapping)
+        if unmatched_count > 0:
+            logger.debug(
+                f"({unmatched_count} legacy weights did not match any parameters in this model component)"
+            )
 
     return mapping