Chart Search Commentary

Optional, user-triggered plain-text AI summary of the real numeric data above. Deliberately a separate, lazy endpoint rather than being bundled into /chart-search -- the default search-and-chart flow costs zero LLM calls; this only runs when the user explicitly clicks "Generate AI commentary". Text-only completion (get_ai_response), not vision -- there's no screenshot in this flow, so there's nothing for a vision model to look at; it's purely writing up numbers that were already computed from real market data.

2026-07-31 fix: this used to hard-force "用繁體中文" (Traditional Chinese) into the prompt regardless of the site's selected UI language -- same backend-hardcoded-language bug already fixed once for the screener prompt (task #317/#327) and once for historical_analog_service.py (task #530). Now accepts the same lang param the sibling /chart-search and /multi-timeframe endpoints on this router already take, and swaps in the shared ai_language_instruction(lang) helper (services/i18n.py) instead of a literal Chinese sentence -- same helper api/chat.py and api/ai_analysis.py's screener prompt already use, so this doesn't invent a second translation mechanism for the same concept.