Adds Nebius AI as a supported LLM provider (OpenAI-compatible inference). Nebius is a hackathon co-sponsor.

Changes

• Added nebius/ vendor prefix to _VENDOR_PREFIX_MAP
• Added _call_nebius() dispatch method using OpenAI client with Nebius base URL
• Env var: NEBIUS_API_KEY
• Tests: 3 new test cases covering prefix routing and model variants

Usage

from kalibr import Router

router = Router(
    goal="research",
    paths=["nebius/meta-llama/Llama-3.3-70B-Instruct", "deepseek-chat"]
)
response = router.completion(messages=[{"role": "user", "content": "..."}])

Supported models

• nebius/meta-llama/Llama-3.3-70B-Instruct
• nebius/Qwen/Qwen2.5-72B-Instruct
• nebius/mistralai/Mistral-Nemo-Instruct-2407