Mark the transforms a dequantized save keeps in the quantizer, not in core loading
update_weight_conversions already receives the model's conversions, so conversion_mapping.py keeps main's code.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>