pub fn normalize_model(model: &str) -> String
Normalize a transcript model id: lowercase, strip provider prefixes, drop the 1M-context tag (pricing is identical to the base model).