Add OrcaRouter as a built-in model provider

[sailt-k]

This PR adds OrcaRouter as a built-in model provider. I'm an engineer on the OrcaRouter team.

OrcaRouter is an OpenAI-compatible meta-provider (similar to OpenRouter) that serves models from many publishers with a single API key. This PR:

• Adds the orcarouter built-in provider (https://api.orcarouter.ai/v1, authenticated via ORCAROUTER_API_KEY), following the same pattern as the existing OpenRouter provider—so the usual provider-selection and fail-over rules apply.
• Adds OrcaRouter provider mappings to the built-in models it currently serves (gpt-4.1 family, gemini-2.5-pro/flash, deepseek v3/r1).
• Adds OrcaRouter's Fusion models (orcarouter/fusion, fusion-mini, fusion-flash) as built-in models. Fusion runs a panel of frontier models in parallel on hard requests and a judge model returns the strongest answer.
• Adds plandex models available --provider <name>: a provider-generic flag that fetches the live list of models a provider is serving from its OpenAI-compatible /models endpoint (with context lengths and pricing when the provider includes them). This works for any directly-reached provider—openrouter, orcarouter, openai, and custom providers—and makes it easy to find model names to use in custom model configs.
• Docs: provider section, built-in model entries, env var, and CLI reference.

Verified with go build/go test across shared/cli/server, plus live catalog fetches against both OpenRouter (343 models, per-token pricing format) and OrcaRouter (167 models, per-million pricing format).

🤖 Generated with Claude Code