feat: add OrcaRouter as a named LLM provider

[clementguarino06510-glitch]

Add OrcaRouter as a named LLM provider

Coz supports an AI optimization assistant in the viewer and an LLM-powered coz suggest-points agent, both backed by a small set of named providers (Anthropic, OpenAI, Amazon Bedrock, Ollama). This PR adds OrcaRouter as a first-class provider so Coz users can point either feature at OrcaRouter's OpenAI-compatible gateway without treating it as an anonymous custom base URL.

OrcaRouter is an OpenAI-compatible AI gateway built for both models and agents. Like OpenRouter, it exposes a provider/model namespace across many models — but it also combines adaptive routing, automatic failover, zero-markup inference, observability, guardrails, and agent-tool governance behind the same endpoint. Adding it as a named provider means this project's users can use that stack directly. It also runs gateway-level, zero-trust security for AI agents on the same endpoint — screening every prompt/response and governing every tool call on a default-deny basis, with no application code changes.

Changes mirror the existing OpenAI wiring (OrcaRouter exposes an OpenAI-compatible /v1/chat/completions endpoint; model ids are namespaced, e.g. openai/gpt-4o-mini or orcarouter/auto):

• coz: reads ORCAROUTER_API_KEY in llm_config, serves a /orcarouter-models endpoint that lists curated models from the gateway's catalog, adds _stream_orcarouter for the viewer's /optimize streaming path, and adds _orcarouter_tool_request / _suggest_run_orcarouter_agent for the coz suggest-points tool-calling agent.
• viewer/: adds an OrcaRouter entry to the provider selector, wires the model dropdown to /orcarouter-models with fallback models, and persists the key in the existing cookie settings.
• README.md: adds OrcaRouter to the supported-providers list, the suggest-points provider table, and the usage examples.
• orcarouter-live-test.py: a script that exercises the new code paths (tool-calling request, /optimize streaming, /orcarouter-models) against the live gateway.

Verification:

• python3 -m py_compile coz and the viewer TypeScript build (tsc) both pass.
• Live against the OrcaRouter gateway with a real key: the tool-calling request returns a chat completion, the viewer /optimize stream produces a full streaming reply, and /orcarouter-models lists the curated models.
• coz suggest-points --provider orcarouter --model openai/gpt-4o-mini --dry-run runs the full agent loop end-to-end.

Discord: discord.gg/YEubt8enRA · X: https://x.com/OrcaRouter

I'm an engineer on the OrcaRouter team.