fix(rl): preserve teleop intervention flag in step_env_and_process_transition

[remete618]

Summary

• Fixes gym_manipulator: Teleop intervention flag gets overwritten to False in step_env_and_process_transition #3227
• Swaps the dict merge order in step_env_and_process_transition so that processed_action_transition[TransitionKey.INFO] takes priority over info from env.step()
• Previously, the env's default is_intervention: False overwrote the teleop processor's is_intervention: True

What changed

Two lines in src/lerobot/rl/gym_manipulator.py (lines 554-555):

# Before (env info overwrites processor info)
new_info = processed_action_transition[TransitionKey.INFO].copy()
new_info.update(info)

# After (processor info overwrites env info)
new_info = info.copy()
new_info.update(processed_action_transition[TransitionKey.INFO])

Test plan

• Verify TeleopEvents.IS_INTERVENTION is True in transition info when teleop processor sets it
• Existing tests pass