spike(learning): PPO path for DroneTarget-v0

[salim4n]

Validate the PPO route for bounded continuous control in DroneTarget-v0.

Acceptance criteria:

• spike identifies whether PPO should be implemented in TS first, Rust Burn, or hybrid;
• rollout batching, advantage estimation, policy/value update and checkpoint shape are documented;
• minimal experiment produces useful metrics even if learning quality is rough;
• decision notes list blockers before full implementation.