fix: remove hardcoded assistant ack after auto_compact to fix 400

[deanbear]

Summary

• auto_compact 压缩后返回了一条硬编码的 assistant 消息，导致 auto_compact 路径中 messages 末尾是 assistant，下一次 API 调用报 400（conversation must end with a user message）
• manual compact 触发后未 return，导致循环继续，同样以 assistant 结尾调用 API

Changes

• 删除 auto_compact 返回值中的假 assistant 消息（s06_context_compact.py、s_full.py）
• manual compact 后加 return，将控制权还给外层 REPL 等待用户输入
• 同步更新三语文档中的代码示例（docs/zh、docs/en、docs/ja）

Test plan

• 运行 python agents/s06_context_compact.py，输入 /compact，确认不再报 400
• 触发 auto_compact（token 超阈值），确认对话可以继续
• 验证 manual compact 后等待用户输入而非继续循环

Additional fix: remove hardcoded assistant acks (s08–s11)

• 在后续的教程中又遇到了相似的 API 错误 400（conversation must end with a user message）
• background-results 和 inbox 注入后紧跟的假 assistant 消息（"Noted background results." / "Noted inbox messages."）一并删除
• 这类消息没有功能价值，连续两条 user 消息在目标模型（claude sonnet 4.6）上正常工作
• 涉及 s08_background_tasks.py、s09_agent_teams.py、s10_team_protocols.py、s11_autonomous_agents.py、s_full.py 及三语文档

> compact: Compressing...
[manual compact]
[transcript saved: /Users/bear/github/learn-claude-code/.transcripts/transcript_1774252544.jsonl]
Traceback (most recent call last):
  File "/Users/bear/github/learn-claude-code/agents/s06_context_compact.py", line 242, in <module>
    agent_loop(history)
    ~~~~~~~~~~^^^^^^^^^
  File "/Users/bear/github/learn-claude-code/agents/s06_context_compact.py", line 203, in agent_loop
    response = client.messages.create(
        model=MODEL, system=SYSTEM, messages=messages,
        tools=TOOLS, max_tokens=8000,
    )
  File "/Users/bear/github/learn-claude-code/.venv/lib/python3.14/site-packages/anthropic/_utils/_utils.py", line 282, in wrapper
    return func(*args, **kwargs)
  File "/Users/bear/github/learn-claude-code/.venv/lib/python3.14/site-packages/anthropic/resources/messages/messages.py", line 996, in create
    return self._post(
           ~~~~~~~~~~^
        "/v1/messages",
        ^^^^^^^^^^^^^^^
    ...<30 lines>...
        stream_cls=Stream[RawMessageStreamEvent],
        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
    )
    ^
  File "/Users/bear/github/learn-claude-code/.venv/lib/python3.14/site-packages/anthropic/_base_client.py", line 1364, in post
    return cast(ResponseT, self.request(cast_to, opts, stream=stream, stream_cls=stream_cls))
                           ~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/bear/github/learn-claude-code/.venv/lib/python3.14/site-packages/anthropic/_base_client.py", line 1137, in request
    raise self._make_status_error_from_response(err.response) from None
anthropic.BadRequestError: Error code: 400 - {'error': {'message': '{"error":{"code":"400","type":"invalid_params","message":"This model does not support assistant message prefill. The conversation must end with a user message."}}. Received Model Group=claude-sonnet-4-6\nAvailable Model Group Fallbacks=None', 'type': 'None', 'param': 'None', 'code': '400'}}

[vercel]

@deanbear is attempting to deploy a commit to the crazyboym's projects Team on Vercel.

A member of the Team first needs to authorize it.