feat: add OrcaRouter as a built-in AI model provider

[jinhaosong-source]

Why are these changes needed?

Disclosure: I'm an engineer on the OrcaRouter team.

OrcaRouter is an OpenAI-compatible meta-router: a single endpoint that fronts 150+ upstream
models and adds server-side routing strategies and an ordered per-request fallback list. On top
of routing it layers a security/governance product (request-level guardrails such as policy
enforcement, PII/secret redaction, and per-key access controls) and a declarative routing DSL.

Users can already connect OrcaRouter today via the generic OpenAI Compatible provider type.
This PR makes it a first-class provider so it shows up in the AI Settings provider list with
its own label, brand icon, and the correct base-URL / model placeholders — the same treatment
OpenRouter already gets — instead of asking users to know it's "OpenAI-compatible" and hand-type
the endpoint.

What changed

Mirrors the existing OpenRouter wiring, but backed by the built-in OpenAI-compatible SDK — so
there is no new dependency (same approach the qwen / zhipu / lingyiwanwu provider types
already use):

• packages/openapi — add ORCAROUTER = 'orcarouter' to the LLMProviderType enum. Validation
  is z.enum(LLMProviderType), so no separate string list needs updating.
• apps/nestjs-backend/src/features/ai/util.ts — map ORCAROUTER to
  createOpenAICompatibleWrapper (the same wrapper used by OPENAI_COMPATIBLE, which forces the
  Chat Completions API and fixes non-standard proxy stream responses), and add ORCAROUTER to the
  getAdaptedProviderOptions case that sets includeUsage: true.
• apps/nextjs-app/.../ai-config/constant.ts — add the provider to LLM_PROVIDER_ICONS and to the
  LLM_PROVIDERS dropdown list (label OrcaRouter, base URL https://api.orcarouter.ai/v1).
• packages/icons — add an OrcaRouter icon component and export it.
• apps/nestjs-backend/.../ai.service.spec.ts — add a unit test mirroring the OpenRouter case
  (verifies the new provider type flows through getModelTags without spurious tag inference).

Verification

• pnpm --filter @teable/openapi typecheck, pnpm --filter @teable/icons typecheck, and the
  nestjs-backend typecheck pass with the changes.
• pnpm --filter @teable/backend test-unit ai.service.spec.ts passes (including the new OrcaRouter
  case).

Live-verified end-to-end against https://api.orcarouter.ai/v1 with a real key:

• GET /models → 200, 172 models (orcarouter/auto, openai/gpt-5.2, etc.).
• non-stream POST /chat/completions (openai/gpt-4o-mini) → 200, correct content + usage.
• streaming POST /chat/completions (stream:true, stream_options.include_usage:true) → 200,
  content streamed across chunks with usage in the final chunk — the exact path
  createOpenAICompatibleWrapper + includeUsage:true handles.
• OrcaRouter routing controls (models fallback list + route) accepted → 200.

Notes

• No new npm dependency: OrcaRouter reuses the built-in OpenAI-compatible client rather than a
  dedicated SDK package (unlike OpenRouter's @openrouter/ai-sdk-provider).
• The included icon is a simple placeholder-style mark; happy to swap in an official SVG if
  preferred.
• AI assistance was used to scaffold the provider (modeled on the existing OpenRouter/Qwen wiring)
  and draft this description; I reviewed and tested the result.

[CLAassistant]

All committers have signed the CLA.

[jinhaosong-source]

👋 Gentle nudge — checks are passing and the PR is ready whenever you get a chance to review. Happy to address anything. Thanks for your time!