Record LLM response attributes and finish the span.
Extracts token usage (input/output/total), response model name,
finish reason, tool calls, and assistant messages from the
LLMResult. Token counts are propagated to parent agent spans
for aggregated usage reporting.
Additional LangChain callback keyword arguments.
