feat: identity PATCH update, alias system, name UNIQUE removal

- Add PATCH /api/v1/identity/:identity_uuid endpoint
- Migration 030: remove name UNIQUE, add tmdb_id index
- TMDb upsert: ON CONFLICT (name) -> ON CONFLICT (tmdb_id)
- get_or_create_identity: pre-check by name
- upload_identity: ON CONFLICT (name) -> ON CONFLICT (uuid)
- Search: include aliases in identity text search
- Add scripts/llm_metadata_enhancer.py
- Add DESIGN/IdentityUpdateAndAliasSystem.md

scripts/llm_metadata_enhancer.py

@@ -0,0 +1,104 @@
+#!/usr/bin/env python3
+"""
+LLM Metadata Enhancer for Momentry Identity
+
+Reads identity name + biography from stdin, calls llama.cpp (Gemma4) to
+produce structured metadata: summary, nationality, profession, aliases (BCP 47 locale-tagged).
+
+Output: JSON to stdout (no extra text).
+
+Usage:
+  echo '{"name": "John Smith", "biography": "..."}' | python3 llm_metadata_enhancer.py
+  python3 llm_metadata_enhancer.py --url http://127.0.0.1:8081 < input.json
+
+Requires:
+  - llama.cpp server running (default: http://127.0.0.1:8081)
+  - pip install requests
+"""
+
+import json
+import sys
+import argparse
+import requests
+
+DEFAULT_API_URL = "http://127.0.0.1:8081"
+
+SYSTEM_PROMPT = """You are a metadata structuring assistant for a media asset management system.
+
+Given an identity name and biography text, produce a structured JSON object with these fields:
+
+- summary: 2-3 sentence summary in the same language as the biography
+- nationality: inferred nationality or null
+- profession: array of inferred professions/titles
+- birth_date: YYYY-MM-DD format if available, else null
+- aliases: array of {locale: "BCP47_tag", name: "translated_name"} objects
+  - Use common well-known translations only (do not fabricate)
+  - Locale tags follow BCP 47 (e.g., en, zh-TW, zh-CN, ja, ko, fr, es, yue, th, ar, ru)
+  - Include at least one alias with locale "en" using the original name
+  - Only include locales where the name has a widely recognized translation
+
+Output ONLY valid JSON. No explanation, no markdown, no extra text."""
+
+
+def call_llm(api_url: str, name: str, biography: str) -> dict:
+    user_prompt = f"Identity name: {name}\n\nBiography:\n{biography}"
+
+    payload = {
+        "messages": [
+            {"role": "system", "content": SYSTEM_PROMPT},
+            {"role": "user", "content": user_prompt},
+        ],
+        "temperature": 0.1,
+        "max_tokens": 1024,
+        "stop": [],
+    }
+
+    resp = requests.post(f"{api_url}/v1/chat/completions", json=payload, timeout=60)
+
+    if resp.status_code != 200:
+        raise RuntimeError(f"LLM API error {resp.status_code}: {resp.text}")
+
+    content = resp.json()["choices"][0]["message"]["content"].strip()
+
+    # Strip markdown code blocks if present
+    if content.startswith("```"):
+        content = content.split("\n", 1)[-1]
+        content = content.rsplit("```", 1)[0]
+    if content.startswith("json"):
+        content = content[4:].strip()
+
+    return json.loads(content)
+
+
+def main():
+    parser = argparse.ArgumentParser(description="Enhance identity metadata via LLM")
+    parser.add_argument("--url", default=DEFAULT_API_URL, help="llama.cpp server URL")
+    parser.add_argument("--input", help="Input JSON file (default: stdin)")
+    args = parser.parse_args()
+
+    source = args.input
+    if source:
+        with open(source) as f:
+            data = json.load(f)
+    else:
+        data = json.load(sys.stdin)
+
+    name = data.get("name", "")
+    biography = data.get("biography", "")
+
+    if not name:
+        result = {"error": "name is required"}
+    elif not biography:
+        result = {"error": "biography is required"}
+    else:
+        try:
+            result = call_llm(args.url, name, biography)
+        except Exception as e:
+            result = {"error": str(e)}
+
+    json.dump(result, sys.stdout, ensure_ascii=False, indent=2)
+    sys.stdout.write("\n")
+
+
+if __name__ == "__main__":
+    main()