refactor: remove scripted file intercept — LLM owns all responses

Previously ab_ai_mail.py intercepted file uploads before reaching the
LLM and responded with a hardcoded clarification template. The LLM had
no involvement in the file upload response.

Changes:
- ab_ai_mail.py: remove _post_file_clarification, _find_pending_attachments,
  _describe_zip, and the two-step pending-attachment lookup. All messages
  (text, files, or both) are dispatched to the agent service immediately.
  Files with no text pass an empty message — the LLM decides what to do.
- upload.py: default message changed from hardcoded receipt instruction
  to '' so the LLM determines intent from file content.
- master_agent._synthesize: always runs through the LLM for both single
  and multi-agent cases — no raw templates reach the user.
- master_system.txt: add FILE UPLOADS routing rule so the LLM knows to
  route receipts to expenses_agent without asking for clarification.

New flow: upload → parse → LLM classifies → agent acts → LLM synthesizes
natural response → user sees it. Zero scripted intercepts.

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

agent_service/agents/master_agent.py

@@ -285,22 +285,22 @@ class MasterAgent:
     async def _synthesize(self, reports, context: MasterContext) -> str:
         if not reports:
             return 'No agent responses received.'
-        if len(reports) == 1:
-            return reports[0].summary or '(no summary)'
-        summaries = chr(10).join(f'{r.agent}: {r.summary}' for r in reports)
-        msg = ('Synthesize these agent reports into one coherent response. '
-               'Business language only. No internal IDs. '
-               'Separate: actions completed, items pending approval, recommendations.'
-               + chr(10) + summaries)
+        summaries = chr(10).join(f'[{r.agent}]\n{r.summary}' for r in reports)
+        instruction = (
+            'You are ActiveBlue AI. Convert the following agent report(s) into a clear, '
+            'natural response for the user. Preserve every factual detail: amounts, dates, '
+            'record IDs, item names, and any action taken. Use plain text only — no JSON, '
+            'no markdown code fences. Write as if speaking directly to the user.'
+        )
         try:
             resp = await self._llm.submit(
-                [{'role': 'system', 'content': 'You are a business intelligence assistant.'},
-                 {'role': 'user', 'content': msg}],
+                [{'role': 'system', 'content': instruction},
+                 {'role': 'user', 'content': summaries}],
                 caller='master_synthesis')
-            return resp.content or summaries
+            return (resp.content or summaries).strip()
         except Exception as exc:
-            logger.warning('_synthesize LLM call failed, falling back to raw summaries: %s', exc)
-            return summaries
+            logger.warning('_synthesize LLM call failed, using raw summary: %s', exc)
+            return reports[0].summary if len(reports) == 1 else summaries
 
     async def _update_memory(self, user_id, message, response, reports, directive_id):
         try: