Downpinned LiteLLM for Router.acompletion typing break

[jamesbraza]

As seen in this CI run, BerriAI/litellm#7594 broke Router.acompletion's typing:

llmclient/llms.py: note: In member "achat" of class "LiteLLMModel":
llmclient/llms.py:621:74: error: Argument 2 has incompatible type
"list[dict[Any, Any]]"; expected
"list[ChatCompletionUserMessage | ChatCompletionAssistantMessage | ChatCompletionToolMessage | ChatCompletionSystemMessage | ChatCompletionFunctionMessage]"
 [arg-type]
    ...ponse = await track_costs(self.router.acompletion)(self.name, prompts)
                                                                     ^~~~~~~

This doesn't actually impact runtime usage, but it does impact type checkers. So as a short-term workaround, I just downpin LiteLLM here.