feat: add LiteLLM as an LLM provider

[RheagalFire]

Summary

Add LiteLLM as an LLM provider, giving OpenTalking access to 100+ LLM providers (OpenAI, Anthropic, Google, Bedrock, Azure, Ollama, etc.) through a single interface.

Set OPENTALKING_LLM_PROVIDER=litellm and use LiteLLM model format in OPENTALKING_LLM_MODEL:

# Anthropic Claude
export OPENTALKING_LLM_PROVIDER=litellm
export OPENTALKING_LLM_MODEL=anthropic/claude-sonnet-4-6
export ANTHROPIC_API_KEY=sk-ant-...

# Google Gemini
export OPENTALKING_LLM_PROVIDER=litellm
export OPENTALKING_LLM_MODEL=gemini/gemini-pro
export GEMINI_API_KEY=...

# LiteLLM proxy
export OPENTALKING_LLM_PROVIDER=litellm
export OPENTALKING_LLM_BASE_URL=http://localhost:4000
export OPENTALKING_LLM_API_KEY=sk-proxy-key
export OPENTALKING_LLM_MODEL=gpt-4o

Changes:

• New opentalking/providers/llm/litellm_chat/ provider with LiteLLMChatClient implementing chat_stream() via litellm.acompletion(stream=True, drop_params=True)
• New opentalking/providers/llm/factory.py with create_llm_client() factory dispatching on llm_provider config
• Updated all 5 LLM client instantiation sites to use the factory
• Optional dep: pip install opentalking[litellm]
• 6 unit tests

Backward compatible - defaults to openai_compatible when OPENTALKING_LLM_PROVIDER is unset.

from opentalking.providers.llm.litellm_chat.adapter import LiteLLMChatClient

client = LiteLLMChatClient(
    base_url="",  # optional - only needed for proxy
    api_key="sk-...",
    model="anthropic/claude-sonnet-4-6",
)
async for chunk in client.chat_stream(messages):
    print(chunk, end="")

Validation

• ruff check src/opentalking/core src/opentalking/server src/opentalking/worker src/opentalking/llm src/opentalking/tts src/opentalking/avatars src/opentalking/events src/opentalking/rtc apps/api apps/unified apps/cli tests
• mypy src/opentalking/core src/opentalking/server src/opentalking/worker src/opentalking/llm src/opentalking/tts src/opentalking/avatars src/opentalking/events src/opentalking/rtc apps/api apps/unified apps/cli --ignore-missing-imports
• pytest apps/api/tests apps/worker/tests tests
• cd apps/web && npm run build

Notes

• litellm is an optional dependency (pip install opentalking[litellm]) - no impact on existing installations
• LiteLLM provider does not require OPENTALKING_LLM_BASE_URL (unlike openai_compatible), since LiteLLM routes via the model identifier prefix (e.g. anthropic/claude-sonnet-4-6)
• drop_params=True enabled by default to handle per-provider parameter differences