Improve OpenAI streaming usage reporting

## What changed OpenClaw now forces stream_options.include_usage on every streaming request to OpenAI and OpenAI-compatible endpoints. The update shipped May 19 2026. Local and custom backends that previously returned zero context usage now report actual token counts.

No user configuration is required. The flag is added automatically in the request body for all streaming calls.

## Why it matters Accurate usage data removes a common blind spot for builders who route traffic between paid APIs and self-hosted models. Without the flag, token tracking stayed broken and bills could jump without clear signals.

The change strengthens OpenClaw's position as a reliable daily driver for mixed-provider setups. It pressures other agent tools to fix the same reporting gap or accept higher support load from confused users.

## How to use it Pull the latest OpenClaw release through the existing CLI command. Restart the agent and send a test streaming prompt to any connected backend. Check the response metadata or provider dashboard for non-zero usage values.

The fix works on current stable versions and requires no YAML edits or new environment variables.

## Watch for Stable usage numbers across Ollama, LM Studio, and custom proxies will confirm the fix holds. A provider that still ignores the flag or returns inflated counts would break the improvement. Expect similar default flags for other streaming options in the next release cycle.