Feat/add litellm provider

[RheagalFire]

What has been done to verify that this works as intended?

15 unit tests pass:

$ pytest tests/test_litellm_model.py -v
15 passed in 6.31s

Covers: init with explicit params / env vars / defaults, missing openai import, query, streaming, None content, auth error (401), rate limit (429), timeout, empty choices, tool-to-function conversion.

6 live E2E tests pass (LiteLLM proxy -> Azure Foundry / Anthropic Claude Sonnet 4.6):

Test	Input	Response	Result
Basic query	"What is 2+2?"	"4"	PASS
Multi-turn conversation	system + user + assistant + followup "What is my name?"	"Your name is Alice."	PASS
Streaming longer response	"Count from 1 to 5"	"1, 2, 3, 4, 5" (streamed in chunks)	PASS
Temperature 0 deterministic	"Capital of Japan?" x2	"Tokyo" both times	PASS
System prompt respected	"Respond in exactly 3 words. Describe the ocean."	"Vast, blue, deep."	PASS
Empty user message	""	Graceful response, no crash	PASS

Why is this the best possible solution? Were any other approaches considered?

Follows the exact same pattern as the existing DeepSeek and vLLM providers - uses the openai SDK pointed at a LiteLLM proxy URL. Zero new dependencies. The alternative was adding litellm as a direct Python SDK dependency, but since MLE-agent already uses the openai SDK for multiple providers, pointing it at a LiteLLM proxy is simpler, lighter, and consistent with the vLLM provider pattern.

How does this change affect users? Describe intentional changes to behavior and behavior that could have accidentally been affected by code changes. In other words, what are the regression risks?

Additive only - no existing code modified except registering the new provider in __init__.py. Existing providers and workflows are completely unaffected. Users opt in by setting platform: LiteLLM in .mle/project.yml.

Changes:

• mle/model/litellm.py - new LiteLLMModel class extending Model with query() and stream()
• mle/model/__init__.py - added MODEL_LITELLM = 'LiteLLM', import, and load_model() branch
• tests/test_litellm_model.py - 15 unit tests

Usage:

# .mle/project.yml
platform: LiteLLM
base_url: http://litellm-proxy:4000/v1
api_key: sk-litellm-key  # optional if proxy has no auth

Do we need any specific form for testing your changes? If so, please attach one.

No. A running LiteLLM proxy is needed for live testing. Unit tests run without one (fully mocked).

Does this change require updates to documentation? If so, please file an issue here and include the link below.

No external documentation updates required. The provider follows the same configuration pattern as all existing providers.

Before submitting this PR, please make sure you have:

• confirmed all checks still pass OR confirm CI build passes.
• verified that any code or assets from external sources are properly credited in comments and/or in
  the credit file.

[RheagalFire]

cc @huangyz0918 @leeeizhang

[huangyz0918]

LGTM, thank you @RheagalFire

[leeeizhang]

many thanks! @RheagalFire