feat(envs): add RoboTwin 2.0 benchmark integration

- RoboTwinEnvConfig with 4-camera setup (head/front/left_wrist/right_wrist)
- Docker image with SAPIEN, mplib, CuRobo, pytorch3d (Python 3.12)
- CI workflow: 1-episode smoke eval with pepijn223/smolvla_robotwin
- RoboTwinProcessorStep for state float32 casting
- Camera rename_map: head_camera/front_camera/left_wrist -> camera1/2/3

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

src/lerobot/processor/env_processor.py

@@ -226,3 +226,29 @@ class IsaaclabArenaProcessorStep(ObservationProcessorStep):
 
     def observation(self, observation):
         return self._process_observation(observation)
+
+
+@dataclass
+@ProcessorStepRegistry.register(name="robotwin_processor")
+class RoboTwinProcessorStep(ObservationProcessorStep):
+    """Passthrough step for RoboTwin observations, casting state to float32.
+
+    RoboTwin observations already arrive in LeRobot format (observation.images.*
+    and observation.state), so this step mainly ensures state dtype is float32.
+    """
+
+    def _process_observation(self, observation):
+        processed_obs = dict(observation)
+        if OBS_STATE in processed_obs:
+            state = processed_obs[OBS_STATE]
+            if hasattr(state, "dtype") and state.dtype != torch.float32:
+                processed_obs[OBS_STATE] = state.float()
+        return processed_obs
+
+    def transform_features(
+        self, features: dict[PipelineFeatureType, dict[str, PolicyFeature]]
+    ) -> dict[PipelineFeatureType, dict[str, PolicyFeature]]:
+        return features
+
+    def observation(self, observation):
+        return self._process_observation(observation)