fix(pytorch): Adapt to new Gym API for compatibility

[jiahui-xin]

This PR introduces changes specifically within the pytorch/ directory to adapt the codebase for newer Gym/Gymnasium API versions.

Problem:
The original code within pytorch/ was incompatible with newer Gym/Gymnasium API versions. This was due to changes in the return values of env.reset() and env.step(), and how states (sometimes tuples) are handled.

Solution:
This PR updates the code within pytorch/ to align with the new Gym API:

• get_action: Now correctly handles tuple state inputs by extracting the main observation.
• env.reset(): Updated to state, _ = env.reset() to match the new API (returns observation and info).
• env.step(): Adapted to process five return values (next_state, reward, terminated, truncated, info) and now uses a combined done = terminated or truncated flag.

These changes ensure the pytorch/ examples/code work correctly with environments using the updated API, such as CartPole.

Testing:

• Tested the changes in the pytorch/ folder with: gym 0.26.2, Python 3.9.13 on macOS 15.4.1 (Sequoia).