Add save files for all 6 benchmark scenarios

[jkbennitt]

Summary

• Adds 5 missing .rws save files to docker/saves/ so all 6 benchmark scenarios can run end-to-end (local and Docker)
• New scripts/create_scenario_saves.py generates saves programmatically from the crashlanded base via XML manipulation — no manual RimWorld dev mode needed
• Clones colonist pawns (remapped thing IDs, hediff/gene loadIDs, cleared social relations) and inserts scenario-specific item stacks

Scenario	Difficulty	Colonists	Items
First Winter	Medium	3 (base)	—
Toxic Fallout	Rough	4 (+Val)	wood, food, steel
Raid Defense	Rough	5 (+Val, Jin)	weapons, armor, steel, wood, components
Plague Response	Rough	5 (+Val, Jin)	medicine, food
Ship Launch	Hard	5 (+Val, Jin)	steel, components, plasteel, gold, uranium

Test plan

Automated (no game needed):

• CI passes (lint, mypy, pytest)
• python scripts/run_benchmark.py --smoke-test --ticks 2 completes all 6 scenarios
• python scripts/create_scenario_saves.py --phase b is idempotent (re-running produces identical files)

Manual (requires RimWorld + RIMAPI):

• Load each save via RIMAPI: curl -X POST http://localhost:8765/api/v1/game/load -H "Content-Type: application/json" -d '{"file_name":"rle_first_winter_v1"}' (repeat for all 5 new saves)
• Verify each save loads without XML corruption errors in the RimWorld log
• Verify colonist count matches expected population (3, 4, or 5)
• Verify cloned colonists (Val, Jin) appear in the colonist bar and are controllable
• Verify spawned items are on the ground near colony center and unforbidden
• Run at least 5 ticks of each scenario to confirm agents can deliberate against the game state (all 6 scenarios, OpenRouter Nemotron 120B paid, ~$0.11 total)

Stretch (Docker):

• docker run with all 6 saves mounted via rimworld-linux + rimapi-mod volumes (port 8766)
• Each save loads correctly inside the container (6/6 PASS, correct colonist counts)

🤖 Generated with Claude Code

[jkbennitt]

All test plan items verified:

Live game (OpenRouter Nemotron 120B paid):

• 6/6 scenarios ran 5 ticks each, all colonists survived
• unforbid_all_items() fix confirmed working (55 items per scenario)
• Scores: crashlanded 0.861, first_winter 0.895, toxic_fallout 0.862, raid_defense 0.848, plague_response 0.885, ship_launch 0.741
• Total cost ~$0.11

Docker:

• Container started with rimworld-linux + rimapi-mod volumes, saves mounted from docker/saves/
• 6/6 saves loaded successfully with correct colonist counts (3/3/4/5/5/5)