Make remaining LLM models configurable (writing-support route, judge/participant)

[kcarnold]

Background

The colleague-eval merge moved the colleague model + reasoning effort into the scenario config (chat.model / chat.reasoningEffort in lib/scenarios.json), read by both app/api/chat/route.ts and the validation pipeline. A few model ids remain hardcoded:

• app/api/writing-support/route.ts:145 — gpt-5.2 (the AI writing suggestions, not the colleague)
• scripts/scenario_design/simulate.ts — participant simulator uses gpt-4o
• scripts/scenario_design/judge.ts — LLM judge uses gpt-4o

Ask

Decide whether these should also be config-driven (and where the config should live), then parameterize them so model choices are centralized rather than scattered as literals.

Notes

• Deferred from the colleague-eval merge (branch merge-colleague-evals) per review — lower priority than the colleague model.