fix: handle OpenAI-style router responses + log unparseable model output

The router's response shape varies by upstream: some backends return
.message.content, some choices[0].message.content, some only top-level
.content. Prefer them in that order. Record the actually-served model
(_shre.model) in the ledger and log a snippet of unparseable output for
diagnosis.

Co-Authored-By: Claude Fable 5 <[email protected]>

src/lib/router.js

@@ -14,9 +14,14 @@ export async function chat({ routerUrl, model, messages, agentId, sessionId, ten
     throw new Error(`router -> ${res.status}: ${text.slice(0, 300)}`);
   }
   const data = await res.json();
-  const content = data?.message?.content ?? data?.content;
+  // Response shape varies by upstream: prefer .message.content, then
+  // OpenAI-style choices, then top-level .content (which can carry an
+  // upstream-down apology — hence last).
+  const content = data?.message?.content
+    ?? data?.choices?.[0]?.message?.content
+    ?? data?.content;
   if (typeof content !== 'string' || !content.trim()) {
     throw new Error('router returned empty content');
   }
-  return content;
+  return { content, servedModel: data?._shre?.model ?? null };
 }