Add Grok 4.6 and tighten subagent guidance

[mgaruccio]

Summary

• Add xai/grok-4.6 to the built-in catalog on the Responses API, matching xAI's published model id, 500k context, image input, and low/medium/high/xhigh reasoning (reasoning cannot be disabled).
• Tighten subagent prompt guidance so parents delegate context-heavy exploration and children return compact evidence.

Test plan

• cd packages/ai && npx tsx ../../node_modules/vitest/dist/cli.js --run test/xai-models.test.ts
• cd packages/coding-agent && npx tsx ../../node_modules/vitest/dist/cli.js --run test/system-prompt.test.ts
• ./prime-agent.sh model list grok-4.6 lists xai grok-4.6 500K 500K yes yes

Note

Add xAI Grok 4.6 model and tighten subagent prompt guidance

• Adds grok-4.6 to the xAI model catalog via the OpenAI Responses API with image input, reasoning enabled, xhigh thinking level support, 500k context/token limits, and pricing of $2/$6 per million input/output tokens.
• Extends subagent guidance in rlm.ts to treat sub-agents as context boundaries, prohibit forwarding full conversation history, and require compact returns via a message or file.
• Adds a child agent doctrine line enforcing compact replies in recursion-disabled contexts.

^{Macroscope summarized 1a1dbd3.}