feat(providers): offer Qwen3.8 Max on Alibaba Coding Plan once Alibaba lists it

## Summary

Qwen3.8 Max (`qwen3.8-max`) is not selectable for **Alibaba Coding Plan (China)** / **Alibaba Coding Plan** connections. This currently matches Alibaba's official plan allowlist, so Maka is behaving as designed — this issue tracks adopting the model **once Alibaba adds it to the Coding Plan**.

## Evidence (checked 2026-08-20)

- Maka's allowlist is `alibabaCodingPlanModelIds` in `packages/core/src/provider-registry.ts`; the code declares Alibaba's Coding Plan docs authoritative for it.
- Alibaba's Coding Plan docs ([CN](https://help.aliyun.com/zh/model-studio/coding-plan) / [Global](https://www.alibabacloud.com/help/en/model-studio/coding-plan)) publish an exact-string allowlist: `qwen3.7-plus`, `qwen3.6-plus`, `kimi-k2.5`, `glm-5`, `MiniMax-M2.5`, `qwen3.5-plus`, `qwen3-max-2026-01-23`, `qwen3-coder-next`, `qwen3-coder-plus`, `glm-4.7` — `qwen3.8-max` is **not** listed.
- The models.dev snapshot (Maka's metadata source) has no `qwen3.8-max` entry under `alibaba-coding-plan-cn` / `alibaba-coding-plan`, so re-running `scripts/sync-model-metadata.mjs` does not surface it either.
- Selecting an unlisted model would fail server-side with `model 'qwen3.8-max' is not supported` per Alibaba's Coding Plan FAQ.

Contrast: **Alibaba Token Plan** officially supports `qwen3.8-max` and was enabled in #3157 — users who need Qwen3.8 Max today can use that access path.

## Adoption checklist (blocked on upstream)

1. Alibaba Coding Plan docs list `qwen3.8-max` as a supported exact model id.
2. models.dev publishes `qwen3.8-max` under `alibaba-coding-plan-cn` / `alibaba-coding-plan`; refresh via `scripts/sync-model-metadata.mjs`.
3. Add `qwen3.8-max` to `alibabaCodingPlanModelIds` in `packages/core/src/provider-registry.ts` (China and global share one list), mirroring the #3157 shape: thinking-effort overrides in `model-metadata.ts`, catalog/thinking/wire-contract tests.

Start by checking the Alibaba Coding Plan CN and global documentation and the models.dev entries for qwen3.8-max. Then read alibabaCodingPlanModelIds in packages/core/src/provider-registry.ts, the related model-metadata.ts logic, and the catalog, thinking, and wire-contract tests. Done means the upstream allowlist and metadata agree and the provider tests cover the new model.

Written by the indexing model from the issue text.