Fix stale rollout success state and preserve LIBERO replay videos

[Anirudh-Kavle]

This PR makes two small improvements to examples/libero/main.py:

1. Fixes stale episode success state after exceptions
2. Prevents replay videos from being overwritten across episodes

1. Reset episode success state
done was not initialized before entering the episode loop, so if an exception occurs before env.step() runs, the previous episode's value carries over. This can cause a failed episode to be incorrectly saved and logged as a success. Fixed by adding done = False at the start of each episode.

2. Preserve replay videos across episodes
Replay videos were previously saved using filenames based only on task description and success/failure suffix, causing later episodes to overwrite earlier ones. This makes failure analysis unreliable, since a later success can overwrite an earlier failure video. Added ep{episode_idx:03d} to the filename so all rollout videos are preserved.

Neither change affects evaluation metrics or rollout behavior.