feat: async invoke

[samuelint]

update multiple dependencies

openai = "^1.35.1" -> "^1.93.0"

langchain = { version = "^0.2.6", optional = true } -> { version = "^0.3.26", optional = true }
langchain-openai = { version = "^0.1.8", optional = true } -> { version = "^0.3.27", optional = true }
fastapi = { version = "^0.111.0", optional = true } -> { version = "^0.115.14", extras = ["standard"], optional = true }
langgraph = { version = "^0.2.16", optional = true } -> { version = "^0.5.0", optional = true }
langchain-anthropic = { version = "^0.1.19", optional = true } -> { version = "^0.3.16", optional = true }
langchain-groq = { version = "^0.1.6", optional = true } -> { version = "^0.3.4", optional = true }

langchain-llamacpp-chat-model = "^0.2.2" -> drop
llama-cpp-python = "^0.3.9" -> new
langchain-community = "^0.2.9" -> "^0.3.26"
langchain-experimental = "^0.0.62" -> "^0.3.4"

---

The code change in langgraph changes the code below.

        # messages_modifier -> prompt
        return create_react_agent(
            llm,
            [],
-           messages_modifier="""You are a helpful assistant.""",
+           prompt="""You are a helpful assistant.""",
        )

# langgraph.graph.graph -> CompiledStateGraph
-from langgraph.graph.graph import CompiledGraph
+from langgraph.graph.state import CompiledStateGraph

The code change in openai changes the code below. But I don't know if this is the right way to do it.

+metadata = {key: str(value) for key, value in metadata.items()}

https://github.com/caffeinism/langchain-openai-api-bridge/blob/patch-1/langchain_openai_api_bridge/assistant/adapter/openai_event_factory.py#L71

agent wrapped with async context manager

Similar idea to #48, but this implementation wraps all functions in an async context manager.

This can be used as before.

    def create_agent(self, dto: CreateAgentDto) -> Runnable:
        ...
        return create_react_agent(llm, tools, prompt=prompt)

Like the idea in the previous PR, we also allow the following

    async def create_agent(self, dto: CreateAgentDto) -> Runnable:
        ...
        return create_react_agent(llm, tools, prompt=prompt)

Additionally, it allows the following code

    @asynccontextmanager
    async def create_agent(self, dto: CreateAgentDto) -> Runnable:
        ...
        yield create_react_agent(llm, tools, prompt=prompt)

    async def create_agent(self, dto: CreateAgentDto) -> Runnable:
        ...
        yield create_react_agent(llm, tools, prompt=prompt)

    @contextmanager
    def create_agent(self, dto: CreateAgentDto) -> Runnable:
        ...
        yield create_react_agent(llm, tools, prompt=prompt)

    def create_agent(self, dto: CreateAgentDto) -> Runnable:
        ...
        yield create_react_agent(llm, tools, prompt=prompt)

This enables connection management by using methods from the context manager.

I've verified that anthropic, openai passes pytest except for groq, which I don't have a key for, and the additional behavior works in my use case, the bind_openai_chat_completion method.