fix(train): fix VF scheduler config and reward model hub push

Add missing peak_lr and decay_lr to DistributionalVFConfig scheduler
preset. Fix push_model_to_hub call for reward models.

src/lerobot/rewards/distributional_value_function/configuration_distributional_value_function.py

@@ -98,6 +98,8 @@ class DistributionalVFConfig(RewardModelConfig):
         return CosineDecayWithWarmupSchedulerConfig(
             num_warmup_steps=500,
             num_decay_steps=50000,
+            peak_lr=3e-4,
+            decay_lr=1e-6,
         )
 
     def validate_features(self) -> None:

src/lerobot/scripts/lerobot_train.py

@@ -741,6 +741,8 @@ def train(cfg: TrainPipelineConfig, accelerator: "Accelerator | None" = None):
             # PEFT only applies when training a policy — reward models use the plain path.
             if not cfg.is_reward_model_training and cfg.policy.use_peft:
                 unwrapped_model.push_model_to_hub(cfg, peft_model=unwrapped_model, dataset_meta=dataset.meta)
+            elif cfg.is_reward_model_training:
+                unwrapped_model.push_model_to_hub(cfg)
             else:
                 unwrapped_model.push_model_to_hub(cfg, state_dict=model_state_dict, dataset_meta=dataset.meta)
             preprocessor.push_to_hub(active_cfg.repo_id)