Hosted RL Entrypoint

[manveerxyz]

Note

Adds first-class hosted RL training support and wiring in the CLI.

• New prime rl commands: run (create/start), models, list, stop, delete, init (generates rl.toml), with table/JSON output and W&B integration
• RL API client: RLClient with list_models, list_runs, create_run (supports wandb secrets/monitoring and team_id/run_config), stop_run, delete_run; models RLModel and RLRun
• Config utilities: BaseConfig.from_sources and load_toml for merging CLI args over TOML (exported via utils.__init__)
• CLI main updates: adds rl to the “Lab” group; reorganizes help panels for Account/Lab/Compute
• Minor: update evals run help text; pretty link output for eval results

^{Written by Cursor Bugbot for commit 084b563. This will update automatically on new commits. Configure here.}