fix: add missing app.py entrypoints for el_farol, shape_example, caching_and_replay

[Vanya-kapoor]

Summary

Adds missing app.py entrypoints to three examples that could not be
launched with solara run app.py. Also fixes a relative import in
el_farol uncovered during the same audit.

Examples fixed:

• el_farol — added app.py + fixed from .agents → from agents
• shape_example — added app.py
• caching_and_replay — added app.py

Not included: mmc_queue — uses mesa.experimental.scenarios which
requires Mesa 4.x. Needs a separate PR targeting the main branch.

Verified (Mesa 3.3.1, Python 3.11)

el_farol       ✓ model initializes, app.py present
shape_example  ✓ model initializes, app.py present
caching_and_replay ✓ model initializes, app.py present

What each app.py does

el_farol — SolaraViz with a line chart of attendance over time.
Sliders for num_agents, crowd_threshold, num_strategies,
memory_size. No space component — this model has no grid.

shape_example — SolaraViz with space component showing agents
as arrowheads. Slider for num_agents.

caching_and_replay — SolaraViz with space component (two agent
types in different colours) and happy-agent line chart. Sliders for
homophily, density, minority_pc, radius.

---

GSoC contributor checklist

Context & motivation

In my audit, I found that these four examples had no app.py. And I looked at each model to understand what visualization it needed before writing anything.

What I learned

I learnt the difference between a model with a grid vs without (el_farol has no space component so the app.py is just a plot), and that mmc_queue is actually a Mesa 4.x example which I only discovered when the import failed when I tried to run it.

Learning repo

🔗 My learning repo: https://github.com/Vanya-kapoor/GSoC-learning-space
🔗 Relevant model(s): https://github.com/Vanya-kapoor/GSoC-learning-space/tree/main/models

Readiness checks

• This PR addresses an agreed-upon problem (missing entrypoints
  identified in systematic audit of all 21 examples on Mesa 3.3.1)
• I have read the contributing guide
• I have performed a self-review
• Another GSoC contributor has reviewed this PR: @
• Models verified locally (import + init checks pass)
• Code formatted (ruff check clean)
• No tests required — entrypoint additions only