Move enable_thinking and reasoning_effort under [sampling]

[JannikSt]

Follow-up to #526. These belong under [sampling] alongside other sampling controls, not at top level.

• RLConfig.enable_thinking / reasoning_effort → SamplingConfig
• Mutual-exclusion validator moves with them
• Run summary shows them in the Sampling section
• Template + tests updated

---

Note

Medium Risk
Medium risk because it changes the RL TOML config schema: existing configs using top-level enable_thinking/reasoning_effort will now fail validation and must be moved under [sampling]. Runtime behavior is otherwise unchanged aside from where values are read and displayed.

Overview
Moves hosted RL reasoning controls (enable_thinking, reasoning_effort) from top-level RLConfig into SamplingConfig under [sampling], including relocating the mutual-exclusion validator.

Updates the generated config template, CLI run summary output (now shown in the Sampling section), and the create_run API payload mapping to read these values from cfg.sampling. Tests are updated to require [sampling] usage and add coverage that top-level reasoning_effort is rejected.

^{Reviewed by Cursor Bugbot for commit 4cc1cda. Bugbot is set up for automated code reviews on this repo. Configure here.}