feat(llm): add Atlas Cloud as an LLM provider

[binyangzhu000-sudo]

Summary

Adds Atlas Cloud as an LLM provider. It is an OpenAI-compatible inference platform (DeepSeek,
Qwen, GLM, Kimi, MiniMax), so it registers through litellm's openai/ prefix with an api_base
override — the same routing the existing Novita and Inception entries use.

ENABLE_ATLASCLOUD=true
LLM_KEY=ATLASCLOUD
ATLASCLOUD_API_KEY=apikey-...
ATLASCLOUD_MODEL=deepseek-ai/deepseek-v4-pro

The registration follows the GROQ block: one LLM_KEY built from ATLASCLOUD_MODEL rather than
a hardcoded model list, so the entry cannot go stale as the platform's catalogue changes.

Registered in all six places a provider lives here

Not just the registry — I grepped ENABLE_GROQ to find every touchpoint:

File	What
skyvern/config.py	ENABLE_ATLASCLOUD, ATLASCLOUD_API_KEY, ATLASCLOUD_MODEL, ATLASCLOUD_API_BASE (default https://api.atlascloud.ai/v1)
skyvern/forge/sdk/api/llm/config_registry.py	the ATLASCLOUD config
skyvern/cli/doctor.py	provider health check — both key and model are required to register, so it checks the pair, the way OPENAI_COMPATIBLE does
docs/developers/self-hosted/llm-configuration.mdx	a section after Groq
docker-compose.yml	commented env block beside the other providers
env.litellm.example / env.ollama.example	ENABLE_ATLASCLOUD=false in the "disable other providers" lists

Validation

Verified through litellm itself, not only against the raw HTTP endpoint — i.e. the exact shape
config_registry.py registers:

litellm.completion(
    model="openai/deepseek-ai/deepseek-v4-pro",
    api_base="https://api.atlascloud.ai/v1",
    api_key=...,
    max_completion_tokens=512,
)
# -> finish_reason=stop, content='ok', 35 tokens

• ruff check and ruff format --check on the three touched Python files: unchanged from the
  pre-branch baseline (2 pre-existing findings, none added — checked by stashing and re-running).
• The docs section notes one caveat: the suggested default is a reasoning model that spends
  completion tokens on a hidden chain of thought, so a small LLM_CONFIG_MAX_TOKENS can return an
  empty response with finish_reason="length". Non-reasoning ids such as deepseek-ai/DeepSeek-V3.1
  are unaffected.

No credentials in the diff.

---

🤝 Partnership & contact

This PR comes from the Atlas Cloud team. Beyond the integration above, we'd love to explore a closer collaboration with Skyvern — for example co-marketing or a featured integration.

If that sounds interesting, reach out anytime:

• 📧 marketing@atlascloud.ai
• 🌐 https://www.atlascloud.ai
• 📚 Docs: https://www.atlascloud.ai/docs

And of course, happy to revise this PR to match your project's conventions — just leave a comment. 🙌