CLI pipeline rate-limit resilience via fallback model pattern

When building CLI pipelines that make multiple LLM calls per item (e.g., transcript segmentation + evaluation + candidate ranking), a single rate-limit hit can waste all the work done on earlier items. A simple fallback pattern:

models_to_try = [primary_model] + ([fallback_model] if fallback_model else [])
for model in models_to_try:
    try:
        result = call_llm(item, model=model)
        break
    except ContentError:
        break  # fallback won't help with bad input
    except Exception as e:
        if model != models_to_try[-1] and 'RateLimitError' in type(e).__name__:
            log(f'rate-limited on {model}; falling back to {models_to_try[-1]}')
            continue
        raise

Key design choices:

• Only fall back on rate-limit errors, not content/auth errors
• Content errors (missing transcript, bad JSON) break immediately — a different model won't fix bad input
• The fallback is per-item, not per-run — if the primary recovers mid-run, items already probed with the fallback stay as-is
• Expose as a CLI flag with a sensible default (--fallback-model) so users can disable it or swap models without code changes