Refactor RL system to integrate with existing grSim simulator and reuse shared constants

[Copilot]

Description

The RL system was fully standalone with its own physics engine and duplicated geometry/physics constants from the C++ codebase. This refactors it to use the existing grSim simulator as its physics backend and source constants from a single shared module that mirrors rj_constants/constants.hpp.

Core changes:

• constants.py (new) — Single Python source of truth for physical constants, mirroring constants.hpp and network.hpp (robot/ball dimensions, field geometry, sim ports, physics params)
• sim_client.py (new) — grSim UDP protobuf client using the same RobotControl/SimulatorCommand/SSL_WrapperPacket protocol as sim_radio.cpp
• config.py — Removed FieldConfig and PhysicsConfig (duplicated geometry); only RL-specific config remains (RewardConfig, TrainingConfig, EnvConfig)
• env.py — Replaced built-in physics with grSim communication (--use-sim); retains lightweight fallback using shared constants for CI/testing
• state.py, action.py, reward.py — Now reference constants module instead of duplicated config values
• proto_gen/ — Generated Python protobuf bindings from existing src/rj_protos/protos/ssl_simulation_*.proto

# Before: duplicated constants
class FieldConfig:
    length: float = 9.0      # duplicates constants.hpp
    robot_radius: float = 0.09  # duplicates constants.hpp

# After: single source of truth
from rj_rl import constants
constants.FIELD_LENGTH      # mirrors rj_constants/constants.hpp
constants.ROBOT_RADIUS      # mirrors rj_constants/constants.hpp
constants.SIM_COMMAND_PORT  # mirrors rj_common/network.hpp

Associated Issue

RL system architecture review feedback — eliminate standalone physics and geometry duplication.

Steps to test

1. cd src/rj_rl && pip install -e . && python -m pytest test/ -v — 52 tests pass
2. python -m scripts.train --timesteps 500 --seed 42 — fallback mode training works
3. Start grSim, then python -m scripts.train --timesteps 5000 --use-sim — simulator-backed training

Expected result: All tests pass; training runs in both fallback and grSim modes; no duplicated physics constants remain in config.py.

---

💡 You can make Copilot smarter by setting up custom instructions, customizing its development environment and configuring Model Context Protocol (MCP) servers. Learn more Copilot coding agent tips in the docs.