pipelineagent: expose timing metrics & api errors wip

[theomonnom]

currently: stt, llm, eou, vad, tts timing info

agent = PipelineAgent(...)

@agent.metrics.on("metrics_collected")
def on_metrics(metrics):
     pass

• This also include an initial pass on exposing errors

[changeset-bot]

🦋 Changeset detected

Latest commit: 135455e

The changes in this PR will be included in the next version bump.

This PR includes changesets to release 11 packages

Name	Type
livekit-plugins-elevenlabs	Patch
livekit-plugins-anthropic	Patch
livekit-plugins-cartesia	Patch
livekit-plugins-deepgram	Patch
livekit-plugins-browser	Patch
livekit-plugins-google	Patch
livekit-plugins-openai	Patch
livekit-plugins-playht	Patch
livekit-plugins-silero	Patch
livekit-plugins-azure	Patch
livekit-agents	Patch

Not sure what this means? Click here to learn what changesets are.

Click here if you're a maintainer who wants to add another changeset to this PR

[davidzhao]

I'm wondering if the general design should be more generic.. instead of different events. Imagine the following interface:

pipeline_agent.on('metrics_updated', lambda metrics: print(metrics))

where a single callback would be able to collect all sorts of metrics. we could make the metrics more consistent too:

@dataclass
class AgentMetrics:
    type: Literal["tts", "llm",...]
    metrics: Union[TTSMetrics, LLMMetrics, STTMetrics, ...]

[davidzhao]

looks good! a few nits.

can we also create an example that shows how one might collect stats? the example could just print them out to console (or perhaps a default console log handler that we can include by default?)