Cassette model picker

This is an explicit settings action. Never invoke it automatically during media ingestion or before an edit. A fresh Cassette session already uses GPT-5.6 Luna with xhigh thinking.

1. Reuse the active Cassette session_id from the current conversation. If no Cassette media session exists yet, tell the user to add media first; do not invent an id or create an edit.
2. Call cassette_config with only session_id to read the current selection and static options.
3. If the invocation already names a valid model and/or thinking level, call cassette_config with those values and confirm the saved choice in one line.
4. Otherwise show the model and thinking choices as two compact numbered lists and wait for the user's selections. Then call cassette_config with the selected model label and thinking value.
5. Explain that the setting is scoped to this Cassette session and applies from the next edit turn.

Selectable models are GPT-5.6 Luna and GPT-5.4 Mini. Thinking values are off, minimal, low, medium, high, and xhigh (display xhigh as “Extra High”).