Add top-level enable_thinking and reasoning_effort to RL config

[JannikSt]

Adds the two hosted RL reasoning controls added to the backend in platform#1390 as top-level fields on RLConfig, so users don't have to guess the nested sampling.extra_body.chat_template_kwargs path.

• enable_thinking: bool — Qwen3.5 / Nemotron
• reasoning_effort: "low" | "medium" | "high" — GPT-OSS
• Mutual exclusion validated client-side, mirrored from the backend schema
• Forwarded 1:1 in create_run; server resolves model-family + trainer-image-version path

---

Note

Low Risk
Low risk: adds optional config fields and forwards them to the existing run-creation API, with a small new validation rule that only errors when both options are set.

Overview
Adds two optional, top-level hosted RL reasoning controls to RLConfig—enable_thinking and reasoning_effort—including client-side mutual-exclusion validation and updated rl init template/docs.

Updates RLClient.create_run and the prime rl run flow to display these settings and forward them 1:1 in the /rft/runs payload, and adds tests covering config loading and rejection when both controls are provided.

^{Reviewed by Cursor Bugbot for commit 2dec5c6. Bugbot is set up for automated code reviews on this repo. Configure here.}

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 43cbb1a. Configure here.}