add curiosity grid exploration example

A new embodied_ai example demonstrating intrinsic curiosity: the agent
keeps a per-cell visit-count map, computes a novelty score for every
free cell, picks the most novel reachable cell as a target, walks the
A* path to that target while committing to it, and repeats. The loop
terminates when the visited coverage of free cells crosses a configurable
threshold.

This differs from frontier exploration (05) in that novelty persists
even after a cell is observed, so the agent keeps revisiting under-
explored areas instead of stopping once UNKNOWN cells are gone.

Includes the runnable example, a smoke test asserting coverage success
and path commitment, and a README GIF.

Also fixes two pre-existing GIF makers: 08_interactive_mpc now unwraps
StepResult via .as_tuple(), and 09_blocked_path_recovery imports
draw_blocked_path_scene from pir.worlds.blocked_path instead of from
the example module.

Author: rsasaki0109