Implement PettingZoo wrapper for OGM

[qu-bit1]

Creates a new PettingZoo-compatible environment in env/pivoting_cubes_env.py to wrap the OccupancyGridMap simulation logic. This provides a standardized interface for training multi-agent reinforcement learning algorithms.

• Implements a multi-agent step() function that supports simultaneous moves
• Adds robust validation to prevent both destination collisions and disconnection

Closes #43