fix(ai): isolate heavy Ollama workloads from GCP alert lane

apps/api/src/services/image_analysis_service.py

@@ -21,7 +21,6 @@ AWOOOI — Image Analysis Service (Phase 34, ADR-067)
 from __future__ import annotations
 
 import base64
-import os
 import time
 from pathlib import Path
 from typing import TYPE_CHECKING
@@ -29,14 +28,13 @@ from typing import TYPE_CHECKING
 import httpx
 import structlog
 
-from src.core.config import get_settings
 from src.services.model_registry import get_model
+from src.services.ollama_endpoint_resolver import resolve_ollama_endpoint
 
 if TYPE_CHECKING:
     pass
 
 logger = structlog.get_logger(__name__)
-settings = get_settings()
 
 # D1 集中化 2026-04-11: 從 models.json providers.ollama.models.image_analysis 讀取
 _MODEL = get_model("ollama", "image_analysis")
@@ -124,7 +122,7 @@ class ImageAnalysisService:
             image_b64 = base64.b64encode(image_path.read_bytes()).decode()
             http = await self._get_http()
             resp = await http.post(
-                f"{settings.OLLAMA_URL}/api/generate",
+                f"{resolve_ollama_endpoint('image_analysis')}/api/generate",
                 json={
                     "model": _MODEL,
                     "prompt": question,