feat(providers): add Bailian provider options

[ewen-poch]

Problem

Bailian's current Qwen 3.8 and related Qwen/DeepSeek catalogs were not available as official providers. The Bailian Token Plan subscription also uses separate fixed endpoints and a smaller, model-specific catalog.

Proposed change

• Add Bailian with China and Global DashScope regions, Anthropic Messages, OpenAI Chat Completions, and Responses support.
• Add the curated Qwen 3.8/3.7/3.6 and DeepSeek V4 catalog, 1M context metadata, and documented image-input capabilities.
• Add Bailian for Plan with its Token Plan Anthropic/OpenAI endpoints, subscription key link, curated Qwen/GLM/DeepSeek catalog, and image-input capabilities. The Plan provider does not advertise Responses support.
• Bundle the LobeHub Bailian color SVG and reuse it for both provider entries.

Sources: Bailian model marketplace, Bailian for Plan model documentation, and LobeHub static Bailian icon.

Scope and non-goals

This is an additive provider-registry and Settings UI change. It does not change persistence, IPC, data relationships, or endpoint-normalization architecture. The existing OfficialModel contract stores context-window size but not separate maximum input, output, or chain-of-thought token limits, so the Qwen 3.8 entry records its 1M context without expanding that data model.

No authenticated live calls were made because validation requires user-owned Bailian keys.

Acceptance criteria and validation

• Provider endpoint, region, model, Responses, multimodal, API-key-link, and Settings icon behavior → npm test -- --run src/shared/provider-registry.test.ts src/renderer/src/pages/settings/ProviderForm.render.test.tsx src/renderer/src/pages/settings/provider-form-value.test.ts → 3 files / 91 tests passed.
• Type safety across main and renderer → npm run typecheck → passed.
• Repository lint rules → npm run lint → passed with 0 errors and 18 pre-existing warnings outside this diff.
• Full regression suite → npm test → 684 files / 10,069 tests passed; six unrelated load-sensitive tests failed under the full concurrent run. Re-running those six files at lower load passed all 379 tests.
• Patch integrity → git diff --check origin/main...HEAD → passed.

Review focus

• Confirm the documented full Anthropic endpoint is correctly normalized by the existing shared URL logic.
• Confirm whole-catalog Responses support for standard Bailian and the absence of Responses support for Bailian for Plan.
• Confirm the curated multimodal lists match the marketplace/Plan capability labels.

[github-actions]

Codex Review

Verdict: mergeable

No actionable findings.

Summary: Static inspection found no concrete merge-blocking defect in the pull request changes. Branch and title prechecks are valid.