feat: add OrcaRouter as a named completion model

[XiaoHuo888-hue]

Description

Adds OrcaRouter as a named registered completion model, plus a small api_key_env resolution in LiteLLMCompletionModel so provider API keys stay in the environment instead of in morphik.toml.

This adds a dedicated OrcaRouter provider rather than relying on the generic OpenAI-compatible endpoint, mirroring how this repo already treats openai_gpt4-1-style aggregators. Named routers such as orcarouter/auto pick an upstream per request. OrcaRouter is an OpenAI-compatible gateway that exposes 150+ models behind one API key, and it also provides gateway-level security controls for AI agents.

Changes

• core/completion/litellm_completion.py — LiteLLMCompletionModel.__init__ now resolves an api_key_env entry in a registered model config (e.g. ORCAROUTER_API_KEY) into the api_key passed to LiteLLM, then removes api_key_env so it isn't forwarded as a LiteLLM kwarg. Models without api_key_env are untouched.
• morphik.toml — new named orcarouter_auto registered model: openai/orcarouter/auto via https://api.orcarouter.ai/v1, reading ORCAROUTER_API_KEY from the environment (keys start with sk-orca-).
• core/tests/unit/test_litellm_completion_orcarouter.py — unit tests covering the resolution (key present, key missing, no-op for other models, unknown key raises).

Usage

# morphik.toml
[completion]
model = "orcarouter_auto"

Checks

• pytest core/tests/unit/test_litellm_completion_orcarouter.py — 4 passed
• ruff check on the changed files — clean
• L3 live test: with a real key, LiteLLMCompletionModel.complete() returns text from https://api.orcarouter.ai/v1 through the resolved key

Disclosure

I'm an engineer on the OrcaRouter team. Happy to adjust naming or scope — or close this if out of scope.