feat: Enable dynamic toggling of tool choice in v3 agent loop for OpenAI [LET-4564] (#5042)

* Add subsequent flag

* Finish integrating constrained/unconstrained toggling on v3 agent loop

* Update tests to run on v3

* Run lint

letta/llm_api/anthropic_client.py

@@ -183,6 +183,7 @@ class AnthropicClient(LLMClientBase):
         llm_config: LLMConfig,
         tools: Optional[List[dict]] = None,
         force_tool_call: Optional[str] = None,
+        requires_subsequent_tool_call: bool = False,
     ) -> dict:
         # TODO: This needs to get cleaned up. The logic here is pretty confusing.
         # TODO: I really want to get rid of prefixing, it's a recipe for disaster code maintenance wise