Update example command for finetuning smolvla with pretrained VLM and action expert intialized from scratch in README

Explicitly adding --policy.load_vlm_weights=true as the pretrained VLM weights seem not to be loaded with the command provided.

README.md

@@ -56,5 +56,6 @@ python lerobot/scripts/train.py \
   --output_dir=outputs/train/my_smolvla \
   --job_name=my_smolvla_training \
   --policy.device=cuda \
+  --policy.load_vlm_weights=true \
   --wandb.enable=true
 ```