[Feat] Add Qwen3.8 Max to OpenCode Go

[zoomote]

​Opened on behalf of @taltas. Follow up by mentioning @roomote, in the web UI, or in Discord.

What changed

• Added native OpenCode Go metadata for qwen3.8-max, including context and output limits, multimodal support, pricing, caching, and streamed reasoning output.
• Routed Qwen3.8 Max through the required Anthropic Messages endpoint.
• Kept LiteLLM's separate OpenAI-compatible reasoning preservation behavior while avoiding an unsupported preservation claim on the direct Anthropic route.
• Added focused registry, pricing, and protocol tests.

Why this change was made

OpenCode Go now serves Qwen3.8 Max. Without curated metadata and routing, Zoo Code can discover the ID dynamically but cannot reliably expose its native limits, calculate costs, or select the correct wire protocol.

Impact

OpenCode Go users can select Qwen3.8 Max with accurate model capabilities and cost reporting. Streamed reasoning remains visible, but it is not advertised as preserved across requests because the gateway does not provide a usable signed-thinking continuity field. Existing models and the default selection are unchanged.

Related PRs

• Zoo-Code-Docs #30

Screenshots

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[zoomote]

Verified the supplied review against the direct Anthropic stream and history paths. The inert preserveReasoning claim was valid and is removed in b826a87; the final diff cleanup is at 8b77b88. LiteLLM keeps its independent OpenAI-compatible preservation allowlist. Signature replay and reasoning controls were not added because the available first-party contract exposes empty placeholder signatures and does not confirm those controls for OpenCode Go/Qwen3.8 Max.

[zoomote]

Resolved the merge conflicts with main in merge commit 33c2257.

Both conflicts were overlapping test additions from the merged GLM-5.3 work and this PR's Qwen3.8 Max work. The resolution preserves both model-specific test cases and keeps the updated GLM-5.3 case-insensitive assertion alongside Qwen3.8 Max. The full test suite, type checks, lint, and diff checks pass.