[Feat] Add GLM 5.3 support

[roomote-community]

​Created by Roomote. Follow up by mentioning @roomote, in the web UI, or in Discord.

What changed

• Added GLM 5.3 to the recommended task model catalog.
• Added verified opencode-go/glm-5.3 and zai-coding-plan/glm-5.3 mappings.
• Updated OpenCode Go and Z.AI Coding Plan to use GLM 5.3 as their default coding model.
• Retained GLM 5.2 recommendations and defaults for providers that serve 5.2 but do not yet expose 5.3, including direct Z.AI.
• Preserved existing GLM 5.2 runtime compatibility for explicitly configured models and snapshot resumes.
• Replaced a mocked OpenRouter GLM 5.3 lookup fixture because OpenRouter's live catalog does not currently list that model.

Why this change was made

GLM 5.3 is available through OpenCode Go and Z.AI Coding Plan and should replace GLM 5.2 for those providers. Other providers continue recommending GLM 5.2 until they expose 5.3. The new model IDs were verified against the live models.dev catalog.

Impact

OpenCode Go and Z.AI Coding Plan users receive GLM 5.3 by default. OpenRouter, Vercel AI Gateway, Requesty, Baseten, Together AI, OpenCode Zen, and direct Z.AI continue to offer GLM 5.2 rather than losing their existing recommendation.

[roomote-community]

No outstanding code issues. See task

• packages/types/src/model-provider-config.ts:745 — Direct Z.AI onboarding again defaults to GLM 5.2, and GLM 5.2 suggestions are restored for other providers despite this PR's stated scope. — dismissed: the clarified PR scope intentionally retains GLM 5.2 wherever GLM 5.3 is unavailable, and the PR description now reflects that behavior.

_{Reviewed deebc6c}

[roomote-community]

Updated in deebc6c2: GLM 5.2 remains recommended and defaulted everywhere it is available but GLM 5.3 is not. Only OpenCode Go and Z.AI Coding Plan advance to GLM 5.3. The provider-split regression test now locks that behavior, all required CI checks pass, and the PR description reflects the final scope. Browser proof was attempted but onboarding remained stuck on a loading state before the model-provider UI.