feat(core): add qwen3.8-max-preview to Alibaba Token Plan

[Lynskylate]

What

Add the new Qwen3.8 Max Preview flagship (released 2026-07-19 on models.dev) to the Alibaba Token Plan catalog, for both the China (cn-beijing) and global (Singapore) access paths. Configuration mirrors the existing qwen3.7/qwen3.6 entries.

Changes

• packages/core/src/provider-registry.ts — register qwen3.8-max-preview at the top of alibabaTokenPlanModelIds (newest flagship first), which feeds fallbackModels for both regions.
• packages/core/src/model-metadata.generated.ts — add generated metadata entries for alibaba-token-plan-cn (/zh docs) and alibaba-token-plan (/en docs): 1M context, 131072 max output, vision + reasoning + function calling. Source: https://models.dev/api.json.
• Tests — update the model-catalog and llm-connections fixtures that pin the exact token-plan model list, including the entries[0] assertions (now qwen3.8-max-preview).

The existing provider-registry validation already enforces that every allowlisted id has functionCalling in the generated snapshot, so the new entry is covered by that guard.

Verification

• npm install ✔
• npm --workspace @maka/core run build ✔
• npm --workspace @maka/core run test:dist ✔ (1132 tests)
• node --test scripts/sync-model-metadata.test.mjs ✔ (30 tests)

Notes

• Hand-edited model-metadata.generated.ts surgically rather than running scripts/sync-model-metadata.mjs, because a full regeneration drifts ~587 lines across unrelated providers (ollama-cloud renames, moonshot additions, lifecycle flips). The added entry exactly matches what the sync script emits for this model.