fix(models): validate MiniMax IDs and sync model docs/runtime

[islee23520]

Summary

• Correct MiniMax M2.7 fallback entries so each provider uses a model ID that exists in its own catalog.
• Remove the nonexistent OpenCode/OpenCode Go highspeed MiniMax fallback path and split direct MiniMax, Vercel, OpenCode, and OpenCode Go IDs.
• Sync runtime fallback chains, tests, CLI snapshots, docs, README files, and examples with the corrected model guidance.
• Carry over the relevant GPT-5.5 documentation alignment from docs: align GPT-5.5 model references #3691 where this branch touches the same model-guide surfaces.
• Correct stale model-cache guidance from opencode models --refresh to opencode models.

Relationship to #3691

• docs: align GPT-5.5 model references #3691 is a docs-only PR focused on non-mini GPT-5.4 -> GPT-5.5 references.
• This PR carries the relevant GPT-5.5 documentation alignment from docs: align GPT-5.5 model references #3691 into the broader model validation branch.
• The primary scope here is independent from docs: align GPT-5.5 model references #3691: provider-specific MiniMax ID validation, runtime fallback-chain updates, tests/snapshots, and CLI cache guidance.
• Because this PR changes runtime model requirements and generated test expectations, it should be reviewed as a model/runtime fix plus documentation sync, not as only a GPT docs cleanup.

Changes

• Direct MiniMax providers use MiniMax-M2.7-highspeed.
• Vercel uses namespaced MiniMax IDs such as minimax/minimax-m2.7-highspeed and minimax/minimax-m2.7.
• OpenCode and OpenCode Go use the available lowercase minimax-m2.7 catalog entry only.
• Explore/Librarian utility chains now distinguish direct MiniMax, Vercel, OpenCode, and OpenCode Go model IDs.
• Atlas, Sisyphus-Junior, Writing, Quick, and Unspecified-low chains use the provider-specific non-highspeed MiniMax fallback where appropriate.
• Docs/examples distinguish GPT-5.5 heavy-reasoning defaults from still-valid GPT-5.4 Mini/Fast/Nano utility models.

Testing

• opencode models | rg "minimax|MiniMax"
• models.dev validation script for all MiniMax provider/model pairs
• bun test src/shared/model-requirements.test.ts src/cli/model-fallback.test.ts
• bun run typecheck
• bun run build
• bun run script/run-ci-tests.ts

Related

• Carries relevant GPT-5.5 doc alignment from docs: align GPT-5.5 model references #3691.
• Closes Docs: realign model guidance after GPT-5.5 runtime promotion #3728 for the GPT-5.5/model-guide documentation drift tracked from this branch.
• Related to feat(opencode-go): upgrade to GLM 5.1 and Kimi K2.6 models and change non-existent minimax-m2.7-highspeed #3595, but this PR is narrower: provider-specific MiniMax ID validation only, without broader Kimi/GLM model upgrades.
• Related to feat: add MiniMax Coding Plan as model provider fallback #2438 for MiniMax provider fallback context.