Revert feet_slide defensive slicing; simplify Digit asset_cfg

The feet_slide change in 07265de added `asset_cfg.body_ids[:contacts.shape[1]]`
to work around an assumed Newton body-name duplication for closed-loop bodies.
Verified empirically: USD has 2 bodies matching `.*_leg_toe_roll` on both
backends; the regex does not over-match `.*_rod_roll` auxiliary joints. The
slicing was unnecessary and crashed with the default `SceneEntityCfg('robot')`
(where body_ids is `slice(None)`) — flagged by review bot.

Fixes:
- rewards.py: restore `body_lin_vel_w[:, asset_cfg.body_ids, :2]`
- digit/rough_env_cfg.py: restore regex `body_names='.*_leg_toe_roll'`
  matching the sensor_cfg pattern; remove misleading comment.

Tested: Digit PhysX 10-iter and Digit Newton 5-iter both run feet_slide
without crashes.

Author: hujc7

source/isaaclab_tasks/isaaclab_tasks/manager_based/locomotion/velocity/config/digit/rough_env_cfg.py

@@ -49,8 +49,7 @@ class DigitRewards:
         weight=-0.25,
         params={
             "sensor_cfg": SceneEntityCfg("contact_forces", body_names=".*_leg_toe_roll"),
-            # Use exact body names to avoid matching extra MuJoCo constraint bodies from ball-joint rods
-            "asset_cfg": SceneEntityCfg("robot", body_names=["left_leg_toe_roll", "right_leg_toe_roll"]),
+            "asset_cfg": SceneEntityCfg("robot", body_names=".*_leg_toe_roll"),
         },
     )
     dof_torques_l2 = RewardTermCfg(

source/isaaclab_tasks/isaaclab_tasks/manager_based/locomotion/velocity/mdp/rewards.py

@@ -84,9 +84,7 @@ def feet_slide(env, sensor_cfg: SceneEntityCfg, asset_cfg: SceneEntityCfg = Scen
     )
     asset = env.scene[asset_cfg.name]
 
-    # Use sensor body count to slice asset velocities — ensures matching dimensions
-    # even when Newton reports duplicate body entries for closed-loop constraints.
-    body_vel = wp.to_torch(asset.data.body_lin_vel_w)[:, asset_cfg.body_ids[:contacts.shape[1]], :2]
+    body_vel = wp.to_torch(asset.data.body_lin_vel_w)[:, asset_cfg.body_ids, :2]
     reward = torch.sum(body_vel.norm(dim=-1) * contacts, dim=1)
     return reward