fix: force JSON output for Ollama intent classification; fix attachment detection

- ollama_backend: add format='json' for 'master' and receipt_parser
  callers so llama3.1:8b returns valid JSON instead of plain English
- ab_ai_mail: add debug logging to trace attachment_ids from Discuss;
  handle file-only messages and clarification look-back flow

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

agent_service/llm/ollama_backend.py

@@ -24,6 +24,11 @@ class OllamaBackend:
                 kwargs = {'model': self._model, 'messages': messages}
                 if tools:
                     kwargs['tools'] = tools
+                # Force structured JSON output for callers that parse JSON responses.
+                # Without this llama3.1:8b returns plain English instead of JSON.
+                _JSON_CALLERS = {'master', 'expenses_agent_receipt_parser'}
+                if caller in _JSON_CALLERS and not tools:
+                    kwargs['format'] = 'json'
                 client = ollama.AsyncClient(host=self._url)
                 try:
                     response = await asyncio.wait_for(client.chat(**kwargs), timeout=self._timeout)