Add verified quick start

[tlebryk]

Summary

• add a Quick start using LibriSpeech clean and Whisper large-v3 published predictions
• show installation, a single-manifest download, and the switch-rate CLI invocation
• clarify that the example is CPU-only and does not download model weights

Why

New users can now run a concrete supported workflow end to end in a few seconds and understand how to substitute their own model predictions.

Validation

• ran the documented workflow locally; switch rate: 0.462 over 42 examples
• pytest -q — 43 passed
• ruff check src tests
• git diff --check