[Feature] Add OrcaRouter as official LLM provider

[zhenjunchen-png]

Summary

Add OrcaRouter — an OpenAI-compatible LLM gateway routing across 40+ upstream providers (OpenAI, Anthropic, Google, DeepSeek, Qwen, Grok, Kimi, MiniMax, z.ai, etc.) — as a Dify model provider plugin.

The plugin exposes 120 models across three Dify model types: LLM (108), Text Embedding (5), and TTS (7), all served through OrcaRouter's unified https://api.orcarouter.ai/v1 endpoint.

Why

OrcaRouter offers three things Dify users currently can't get from any single upstream:

1. Adaptive routing — A LinUCB contextual bandit picks the optimal upstream per request based on prompt length, complexity, and learned reward (cost + latency + reliability). The same orcarouter/auto model serves cheap summarization and premium code-refactor requests with zero client-side dispatch logic.
2. Fallback chains via extra_body — Per-request fallback model list bypasses upstream outages without changing application code.
3. Pay below list price — Users get the same OpenAI/Anthropic/Google models at OrcaRouter's negotiated rates, with a dashboard showing real-time savings.

Aggregator plugins are already first-class citizens in this repo (OpenRouter, CometAPI, AIHubMix, Novita, SiliconFlow, Together, Fireworks, Groq), so this proposal fits the established pattern.

Companion PR

Implementation in PR #3103.

Disclosure

I'm an engineer on the OrcaRouter team.

[zhenjunchen-png]

Migrated to the community plugins repository per maintainer guidance.

• New PR: Add OrcaRouter plugin dify-plugins#2413
• Plugin source repository: https://github.com/zhenjunchen-png/dify-plugin-orcarouter

Closing this issue; further discussion can continue on the new PR.