Model and training upgrades

[msosav]

This pull request includes several significant updates to the project, focusing on adding a GitHub Actions workflow for formatting, enhancing the gym.py configuration, updating memory address handling, and improving the preprocessing environment setup.

GitHub Actions Workflow:

• .github/workflows/formatting.yml: Added a new GitHub Actions workflow to automate code formatting using Black and Isort on pull requests to the main branch.

Configuration Enhancements:

• config/gym.py: Introduced new configuration parameters (action_freq, exploration_reward, reward_scale, game_with_sound) and updated the __init__ method to use these parameters. Added new valid and release actions, and modified the observation space to include additional game state information.
• config/gym.py: Refactored the step method to use a new run_action method, which handles sending input and ticking the emulator.

Memory Address Updates:

• config/memory_addresses.py: Added detailed comments and new memory addresses for various game states, such as destination data, inventory items, dungeon item flags, and rupees.

Preprocessing Environment:

• utils.py: Replaced the grayscale observation wrapper with a custom DictGrayScaleObservation class to handle dictionary-based observation spaces.

Model Training Configuration:

• main.py: Updated the PPO model configuration for training, including changes to the policy type, learning rate, and other hyperparameters.