Training adapter: SkyRL (trainer-side)

[larstalian]

Bridge OpenRange episodes into SkyRL's training stack so SkyRL can train against OpenRange worlds. This is a trainer-side adapter on the #243 seam — not environment-standard conformance.

Note: SkyRL-Gym (Gymnasium BaseTextEnv) and SkyRL-Agent (OpenAI-style function-call tools) are different contracts, not one interface. The adapter picks whichever touchpoint fits; it must not assume they're the same.

Scope

• Wrap an admitted OpenRange Snapshot + episode so SkyRL's trainer can roll out against it; map EpisodeResult → reward via Add per-domain RewardAdapter #199.
• Decide the touchpoint: SkyRL-Gym BaseTextEnv vs SkyRL-Agent tool surface.
• Optional extra; no SkyRL deps in core.

Acceptance

• An OpenRange world trainable from SkyRL through the adapter, no core changes.
• One end-to-end example + short doc.