Add a means to optionally defer function call completion

[markbackman]

Please describe the changes in your PR. If it is addressing an issue, please reference that as well.

This was motivated by the following Flows issue: pipecat-ai/pipecat-flows#67

The issue is that two consecutive completions occurs in Flows: one for the function call completing and one for the subsequent node's message. This change adds the flexibility to defer the completion for a function call. For the Flows case, this will allow a completion to occur after both the function call and the node's message are added to the context.

This maintains backwards compatibility and attempts to avoid complexity. I'm open to suggestions of different ways to accomplish this. In my testing, this works for Flows and maintains backwards compatibility.

[aconchillo]

LGTM!

[hou-man]

Hey @markbackman thanks for pushing this improvement. Quick feedback that it seems to break function calling for openai realtime pipelines:

                self._function_call_result = None
run_llm
2025-01-21 17:01:44.375 | ERROR    | pipecat.services.openai_realtime_beta.context:_push_aggregation:223 - Error processing frame: 'FunctionCallResultFrame' object has no attribute 'run_llm'

[markbackman]

@hou-man thanks, you're right. I created #1063. We'll fix this in the next release.