fix(ai): increase timeout to 60s, keep model resident in RAM for 24h, and warm up on startup

app/main.py

@@ -35,7 +35,7 @@ from app.reader import (
     fetch_and_parse_rfc, parse_rfc_text, fetch_and_parse_wikipedia,
     extract_wikipedia_title_and_lang
 )
-from app.ai import translate_sentence_ai, get_ollama_status
+from app.ai import translate_sentence_ai, get_ollama_status, warmup_ollama_model
 from app.phrasal import detect_phrasal_verb
 from app.cefr import (
     get_cefr_level_summaries,
@@ -68,6 +68,7 @@ from app.miner import run_mining_task_background
 async def lifespan(app: FastAPI):
     init_db()
     ensure_admin_user()
+    asyncio.create_task(warmup_ollama_model())
     yield
     await close_http_client()