Add OrcaRouter as a named LLM provider

[XiaoHuo888-hue]

Adds OrcaRouter as a named LLM provider, wired the same way the existing
OpenRouter provider is — so the provider registry, model selection and docs
stay consistent.

What's in the change

• New OrcaRouterProvider in crates/chidori/src/providers/orcarouter.rs — a
  thin wrapper over the OpenAI-compatible client pointed at
  https://api.orcarouter.ai/v1/chat/completions.
• ProviderRegistry::from_env() registers it whenever ORCAROUTER_API_KEY is
  set, ahead of the OpenRouter fallback, so an explicit OrcaRouter key wins
  over the zero-setup OAuth sign-in.
• to_orcarouter_slug() translates Chidori model ids to OrcaRouter's catalog
  ids on the way out: claude-sonnet-4-6 → anthropic/claude-sonnet-4.6,
  auto → orcarouter/auto, gpt-4o → openai/gpt-4o, and explicit
  catalog ids (anything containing /) pass through untouched.
• ORCAROUTER_API_KEY is surfaced wherever a provider key is documented
  (CLI demo, init help text, README, docs).

This registers OrcaRouter the same way the
existing OpenRouter provider is wired, so the provider registry, config
reference and docs stay consistent. OrcaRouter is an OpenAI-compatible
gateway: one ORCAROUTER_API_KEY (keys start with sk-orca-) unlocks 150+
models from OpenAI, Anthropic, Google, DeepSeek, Qwen, MiniMax and xAI behind
a single https://api.orcarouter.ai/v1 endpoint. 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.

Verification

• cargo test -p chidori --lib providers::orcarouter — 5/5 pass (new slug
  translation tests).
• Full cargo test -p chidori --lib: 445 pass / 29 fail on this branch vs
  440 pass / 29 fail on main — the 29 failures are byte-for-byte identical
  on both, pre-existing Windows temp-path issues in host_actor,
  host_branch and typescript::loader tests that this PR does not touch.
  The PR adds 5 new passing tests and no new failures. (Three replay tests
  additionally fail only when an ANTHROPIC_MODEL env var is set in the
  shell — this dev session runs via a gateway; with it unset they pass.)
• Live against https://api.orcarouter.ai/v1 with a real key:
  GET /models → 200 (189 models); chat completion on
  anthropic/claude-sonnet-4.6 → 200; orcarouter/auto smart routing → 200;
  invalid key → 401.
• cargo fmt --check and clippy are clean on the changed files (the repo
  already has unrelated fmt diffs and three clippy warnings on main head).

I'm an engineer on the OrcaRouter team.