[POST]
summary = "模拟 AI 流式对话（SSE，逐 token 输出）"
docstring = "流式端点示例：@rpc_method 返回生成器，框架逐块多回复推送，网关以 text/event-stream 转发。curl -N 可观察逐块到达。"
auth_required = false
stream = true
# 流式超时模型: timeout = 流总寿命上限（zenoh query deadline），
# stream_idle_timeout = 网关侧连续无数据块的最大间隔
timeout = 300.0
stream_idle_timeout = 10.0
params = [
    "prompt: str",
    "tokens: int"
]