Add Dify integration and enhance workflow node specifications

- Introduce new fields `dify_api_url` and `dify_api_key` in `AssistantConfig` for Dify API integration.
- Update `requirements.txt` to include `dify-client-python` for Dify SDK support.
- Modify `config_resolver` to handle Dify connection information.
- Add a new `globalNode` type in workflow specifications to provide unified settings across workflows.
- Enhance node specifications with additional constraints and default values for better configuration management.
- Update frontend components to support the new `globalNode` type and its properties, improving workflow editor functionality.

backend/services/brains/base.py

@@ -1,46 +1,117 @@
-"""「大脑」抽象:把不同助手类型(prompt/workflow/fastgpt/…)在运行时的差异收口。
+"""Conversation-brain contracts shared by every assistant type.
 
-cascade 管线骨架对所有类型一致(STT → LLM 槽 → TTS),变化的只有:
-  - 谁产出助手文本(LLM 槽里塞什么)——build_llm
-  - 开场白来源(静态 / 外部异步拉取)——greeting
-  - 对话上下文归谁维护——spec.owns_context
-  - 是否支持 realtime——spec.supported_runtime_modes
-
-阶段 1 只抽到「够 fastgpt 用」的程度;workflow 编排仍内联在 pipeline.py,
-待阶段 2 再搬进 WorkflowBrain 收口。
+Brain selects who owns reasoning and conversation state.  The Pipecat pipeline
+still owns media transport, STT/TTS, transcript delivery, and interruption
+semantics.  This keeps assistant-specific orchestration out of pipeline.py
+without coupling brains to Pipecat internals more than necessary.
 """
 
 from __future__ import annotations
 
+from collections.abc import Awaitable, Callable
 from dataclasses import dataclass
-from typing import Protocol, runtime_checkable
+from typing import Any, Protocol, runtime_checkable
 
 from models import AssistantConfig
+from pipecat.adapters.schemas.function_schema import FunctionSchema
+from pipecat.frames.frames import Frame
 from pipecat.processors.aggregators.llm_context import LLMContext
 from pipecat.processors.frame_processor import FrameProcessor
 
 
 @dataclass(frozen=True)
 class BrainSpec:
-    """类型元数据。单一来源,供运行时门控与上下文归属决策复用。"""
+    """Static capabilities used by validation and runtime dispatch."""
 
     type: str
     supported_runtime_modes: frozenset[str]
-    # True:由本服务维护 LLMContext(prompt/workflow);
-    # False:上下文/知识库/工具由外部服务端接管(fastgpt/dify),本地不写 context。
+    # False means context, knowledge bases, and tools live on an external agent.
     owns_context: bool
 
 
-@runtime_checkable
-class Brain(Protocol):
-    """每通电话 new 一个实例(可持有 chatId / 当前节点等会话状态)。"""
+class CallEndPort(Protocol):
+    """Small call-lifecycle surface available to a brain."""
+
+    @property
+    def ending(self) -> bool: ...
+
+    def begin(self, reason: str) -> None: ...
+
+    def arm_after_speech(self) -> None: ...
+
+    async def finish(self) -> None: ...
+
+
+@dataclass(frozen=True)
+class BrainRuntime:
+    """Pipeline-owned capabilities injected into one brain session."""
+
+    context: LLMContext
+    llm: Any
+    queue_frame: Callable[[Frame], Awaitable[None]]
+    set_system_prompt: Callable[[str], None]
+    set_tools: Callable[[list[FunctionSchema] | None], None]
+    call_end: CallEndPort
+
+
+class BaseBrain:
+    """No-op lifecycle defaults for brains without local orchestration."""
 
     spec: BrainSpec
 
     async def greeting(self, cfg: AssistantConfig) -> str:
-        """开场白。内部类型通常直接用 cfg.greeting;外部类型异步拉取后端配置。"""
-        ...
+        return cfg.greeting
+
+    def system_prompt(self, cfg: AssistantConfig) -> str:
+        return cfg.prompt if self.spec.owns_context else ""
 
     def build_llm(self, cfg: AssistantConfig, context: LLMContext) -> FrameProcessor:
-        """返回丢进管线 LLM 槽位的帧处理器(标准 LLMService 或外部托管的伪 LLM)。"""
-        ...
+        raise NotImplementedError
+
+    async def setup(self, cfg: AssistantConfig, runtime: BrainRuntime) -> None:
+        """Register tools and initialize per-call orchestration."""
+
+    async def on_connected(self) -> None:
+        """Handle a connected client after the common greeting is queued."""
+
+    def record_user_message(self, content: str) -> None:
+        """Observe a committed user message for brain-owned routing state."""
+
+    async def on_assistant_text_start(self, turn_id: str) -> None:
+        """Observe the start of a generated assistant turn."""
+
+    async def on_assistant_text_end(
+        self,
+        turn_id: str,
+        content: str,
+        interrupted: bool,
+    ) -> None:
+        """Observe the completion of a generated assistant turn."""
+
+
+@runtime_checkable
+class Brain(Protocol):
+    """One instance per call; implementations may keep conversation state."""
+
+    spec: BrainSpec
+
+    async def greeting(self, cfg: AssistantConfig) -> str: ...
+
+    def system_prompt(self, cfg: AssistantConfig) -> str: ...
+
+    def build_llm(self, cfg: AssistantConfig, context: LLMContext) -> FrameProcessor: ...
+
+    async def setup(self, cfg: AssistantConfig, runtime: BrainRuntime) -> None: ...
+
+    async def on_connected(self) -> None: ...
+
+    def record_user_message(self, content: str) -> None: ...
+
+    async def on_assistant_text_start(self, turn_id: str) -> None: ...
+
+    async def on_assistant_text_end(
+        self,
+        turn_id: str,
+        content: str,
+        interrupted: bool,
+    ) -> None: ...