fix(runTools): sanitize replayed messages

[HAYDEN-OAI]

• I understand that this repository is auto-generated and my pull request may not be merged

Changes being requested

• sanitize assistant messages before runTools replays them in follow-up requests
• preserve enriched runner.messages, events, and parsed completions for callers while removing helper-only parsed and parsed_arguments, null refusal, and other response-only assistant fields from outbound payloads
• update the streaming and non-streaming runner expectations so the regression stays covered

Additional context & links

Fixes #1161.

runTools currently stores parsed completion messages in its history and then sends that history back unchanged on the next tool-loop request. Azure and other strict-compatible endpoints reject the helper-only fields with errors such as Additional properties are not allowed (parsed, refusal were unexpected).

The fix sanitizes only the outbound copy, so existing parsed runner results remain available to users.

Verification:

• pnpm test --runInBand tests/lib/ChatCompletionRunFunctions.test.ts
• pnpm lint
• pnpm test --runInBand

[openai-sdks]

OkTest Summary

❌ Failed for Node SDK PR #1990.

View OkTest run #29280408657

_{SDK merge (2285fd79ad95) · head (b89ac7051e98) · base (ec2f57fd0d66) · OkTest (cb7f0514ccd2)}