feat: add LiteLLM as model backend for 100+ LLM providers

[RheagalFire]

Summary

• Add LiteLLM as a model backend (litellm) using the litellm Python SDK, enabling access to 100+ LLM providers (Anthropic, Bedrock, Vertex, Gemini, Cohere, Mistral, etc.) without needing a separate proxy
• Overrides _post_with_retries to call litellm.acompletion() directly instead of httpx, while inheriting the base class retry logic
• Uses drop_params=True for cross-provider compatibility

Changes

• src/webwright/models/litellm_model.py - new LiteLLMModel backend using litellm SDK with acompletion()
• src/webwright/models/__init__.py - registered "litellm" in _MODEL_MAPPING

Tests

from webwright.models import get_model

model = get_model({
    "model_class": "litellm",
    "model_name": "anthropic/claude-sonnet-4-6",
})
response = model([{"role": "user", "content": "What is 2+2? Answer in one word."}])

SDK Response: 'Four'
SDK E2E PASSED

Example usage

from webwright.models import get_model

# Anthropic via litellm SDK (reads ANTHROPIC_API_KEY from env)
model = get_model({"model_class": "litellm", "model_name": "anthropic/claude-sonnet-4-6"})

# OpenAI via litellm SDK (reads OPENAI_API_KEY from env)
model = get_model({"model_class": "litellm", "model_name": "gpt-4o"})

# Bedrock via litellm SDK (reads AWS creds from env)
model = get_model({"model_class": "litellm", "model_name": "bedrock/claude-3.5-sonnet"})

Risk / Compatibility

• Additive only. Existing OpenAI/Anthropic/OpenRouter backends untouched.
• litellm is an optional dependency, imported at model init time.

[RheagalFire]

@microsoft-github-policy-service agree