Document Atari wrapper reset behavior

[Copilot]

This PR addresses issue #666 by documenting the confusing behavior of env.reset() in Atari environments when using the default terminal_on_life_loss=True setting.

closes #666

Problem

When using make_atari_env() with default settings, the EpisodicLifeEnv wrapper causes env.reset() to perform a no-op step instead of truly resetting the environment when the episode ends due to a life loss (not game over). This leads to unexpected behavior where consecutive calls to reset() return the same observation:

from stable_baselines3.common.env_util import make_atari_env

env = make_atari_env('BreakoutNoFrameskip-v4', n_envs=1, seed=0)

# These two calls may return different observations
obs1 = env.reset()  # First reset
obs2 = env.reset()  # Second reset - same obs

Solution

This PR adds comprehensive documentation to explain:

1. Root cause: The EpisodicLifeEnv wrapper's reset() method only performs a true reset when self.was_real_done is True (game over), otherwise it executes a no-op step.

2. Workaround: Users can avoid this behavior by setting terminal_on_life_loss=False:

env = make_atari_env(
    "BreakoutNoFrameskip-v4",
    n_envs=1,
    wrapper_kwargs=dict(terminal_on_life_loss=False)
)

Changes Made

• Added detailed documentation to EpisodicLifeEnv class and reset() method docstrings
• Updated make_atari_env() docstring with a note about this behavior
• Added a note in the RL Tips documentation with a practical example
• Updated changelog to record this documentation improvement

This is purely a documentation change with no code behavior modifications, ensuring existing functionality remains intact while helping users understand and work around this potentially confusing behavior.

---

💬 Share your feedback on Copilot coding agent for the chance to win a $200 gift card! Click here to start the survey.