Fix UTF-8 decode error for providers returning invalid encoding (Ollama)

Some providers (e.g. Ollama) may return response bytes that aren't valid
UTF-8. json.loads() raises UnicodeDecodeError in that case. Add an httpx
response event hook that replaces invalid bytes before the SDK parses the
response body.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

app/ai/service.py

@@ -7,6 +7,7 @@ from pathlib import Path
 import tempfile
 from typing import TYPE_CHECKING, Any, Awaitable, Callable
 
+import httpx
 from openai import AsyncOpenAI
 
 from app.ai.image_understand import (
@@ -28,6 +29,15 @@ log = logging.getLogger(__name__)
 ProgressCallback = Callable[[str], Awaitable[None] | None]
 
 
+async def _sanitize_response_encoding(response: httpx.Response) -> None:
+    await response.aread()
+    raw = response.content
+    try:
+        raw.decode("utf-8")
+    except UnicodeDecodeError:
+        response._content = raw.decode("utf-8", errors="replace").encode("utf-8")
+
+
 @dataclass(slots=True)
 class AIResponse:
     text: str
@@ -103,6 +113,9 @@ class AIService:
                 api_key=provider.api_key,
                 base_url=provider.base_url,
                 timeout=provider.timeout,
+                http_client=httpx.AsyncClient(
+                    event_hooks={"response": [_sanitize_response_encoding]},
+                ),
             )
             self._clients[provider.name] = client
         return client