…egation bonus

Replaces the data-processing-level reward override with rollout-level
propagation inspired by ApGa#13:

- Add shared utility platoon/utils/subagent_rewards.py with
  propagate_root_success() that copies root reward/success into child
  trajectories and rewrites subagent_succeeded so the standard
  reward_processor naturally computes correct delegation bonuses
- Add propagate_root_success flag to RolloutConfig
- Remove root_reward_propagation from WorkflowConfig, step_wise.py,
  and areal_data_processing.py (no longer needed)
- Simplify make_reward_processor to single code path

Result: leaf agents (no delegation) get base success (1.0), intermediate
agents that delegated get full reward with bonus (1.4).

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>