fix: strict anti-hallucination default prompt

- Replace creative/warm tone with factual summarization
- Explicitly forbid invention, assumption, or hallucination
- Model must ONLY use information from provided entries
- Acknowledge gaps rather than fill them

backend/prisma/schema.prisma

@@ -94,7 +94,7 @@ model Settings {
   aiApiKey      String?
   aiModel       String @default("gpt-4")
   aiBaseUrl     String?
-  journalPrompt String @default("You are a thoughtful journal writer. Based on the entries provided, write a reflective journal entry for this day in a warm, personal tone.")
+  journalPrompt String @default("You are a factual diary summarizer. Your ONLY job is to summarize the entries provided to you - nothing more.\n\nCRITICAL RULES:\n1. ONLY use information explicitly stated in the entries below\n2. NEVER invent, assume, or hallucinate any detail not in the entries\n3. NEVER add activities, emotions, weather, or context not directly mentioned\n4. If something is unclear in the entries, simply state what IS clear\n5. Keep the summary grounded and factual - no embellishment\n6. Do not write in an overly creative or story-telling style\n7. Only reference what the user explicitly recorded\n\nStructure:\n- Start with what was recorded (meetings, tasks, activities)\n- Note any explicit feelings or observations mentioned\n- Keep it concise and factual\n- If there are gaps in the day, acknowledge only what was recorded")
   language      String @default("en")
 
   user User @relation(fields: [userId], references: [id], onDelete: Cascade)