docs: show opt-in EP updater in workspace

[Jammy2211]

Summary

• document that expectation-propagation optimisation is undamped by default
• show how to opt in with af.SimplerUpdater(delta=...)
• regenerate the canonical notebook from the updated tutorial script
• keep the workspace change gated on the upstream PyAutoFit PR merging first

Scripts Changed

• scripts/features/expectation_propagation.py — documents the undamped default and demonstrates the optional af.SimplerUpdater
• notebooks/features/expectation_propagation.ipynb — regenerated canonical notebook output

Upstream PR

• feat: expose opt-in EP updater through declarative optimise PyAutoFit#1457

Test Plan

• direct execution of the changed tutorial script: passed
• notebook matches canonical PyAutoHands generation exactly
• PyAutoFit suite: 1667 passed, 6 skipped
• downstream smoke scripts:
  • AutoFit: 8/8 passed
  • AutoGalaxy: 13/13 passed
  • AutoLens: 35/35 passed
  • Euclid: 6/6 passed
• human-acknowledged environment limitations:
  • exact Heart gate reports RED because PyAutoFit is intentionally on feature/ep-optimise-updater, not main
  • notebook kernels cannot start in this sandbox: ZMQError: Operation not permitted (addr='ipc://kernel-ipc-1')
  • the broad AutoLens integration runtime ended with fatal library error, lookup self

Merge only after the upstream PyAutoFit PR is merged.

Generated with ChatGPT Codex.