fix: аудит — 19 фиксов безопасности, надёжности, UI и 16 новых тестов

- S4: bump jinja2>=3.1.4, python-multipart>=0.0.18, httpx>=0.28.0
- S5: _detect_ecosystem — DEFAULT_ECOSYSTEM для неизвестных форматов
- S6: harvester — log.exception() вместо log.error()
- S8: _scan_component — urlencode параметров
- P1: scanner — proc.kill() при таймауте
- P3: api_packages — selectinload(Scan.findings), убран N+1
- P4+P5: утечка _url_locks и _llm_locks при early return
- P6: DB reaper — сброс {'status':'analyzing'} при старте
- UI: htmx-пагинация, фильтры не теряют flagged, 404 с layout
- UI: мобильные таблицы overflow-x, полная стата на дашборде
- UI: i18n статусов в _status_badge, urlencode package_name
- 16 новых тестов: analyze endpoint (6), scanner errors (4),
  webhook signature (2), llm client (4)

guarddog_nexus/core/llm.py

@@ -23,11 +23,7 @@ def _build_user_message(finding: dict) -> str:
     location = finding.get("location", "")
     code = finding.get("code", "")
 
-    prompt = (
-        f"Rule: {rule}\n"
-        f"Severity: {severity}\n"
-        f"Message: {message}\n"
-    )
+    prompt = f"Rule: {rule}\nSeverity: {severity}\nMessage: {message}\n"
     if location:
         prompt += f"Location: {location}\n"
     if code:
@@ -66,9 +62,7 @@ async def analyze_finding(finding_data: dict) -> dict | None:
 
     try:
         async with _llm_semaphore:
-            async with httpx.AsyncClient(
-                timeout=config.llm_timeout, headers=headers
-            ) as client:
+            async with httpx.AsyncClient(timeout=config.llm_timeout, headers=headers) as client:
                 resp = await client.post(url, json=payload)
                 resp.raise_for_status()
                 body = resp.json()