feat: add LiteLLM as LLM provider

[RheagalFire]

[## Summary

Adds LiteLLM as a new LLM provider option, enabling access to 100+ LLM backends (OpenAI, Anthropic, Google, Azure, AWS Bedrock, Ollama, etc.) through a single LiteLLM proxy server.

LiteLLM appears as a selectable "LiteLLM" option in the Django admin provider dropdown alongside the existing providers.

How it works

LiteLLM runs as an OpenAI-compatible proxy, so the implementation reuses pydantic-ai's existing OpenAIModel/OpenAIProvider with a custom base_url pointing at the LiteLLM proxy. No new dependencies are required.

Configuration (Django admin SiteConfig):

• LLM Provider: select "LiteLLM"
• LLM API Base: your LiteLLM proxy URL (e.g. http://localhost:4000/v1)
• AI Model: any model your proxy serves (e.g. anthropic/claude-sonnet-4-20250514)
• LLM API Key: the proxy's virtual key (optional -- local proxies typically run without auth)

Changes

File	What
linkedin/llm.py	_build_litellm() builder + registration in _PROVIDER_BUILDERS + API key validation skip for litellm
linkedin/models.py	LITELLM = "litellm", "LiteLLM" added to SiteConfig.LLMProvider choices
linkedin/migrations/0008_...	Django migration adding the new choice to the llm_provider field
tests/test_litellm_provider.py	11 unit tests covering builder, validation, edge cases

Testing

Unit tests (11 passing):

• Provider choice exists with correct label
• Builder creates OpenAIModel with correct model name
• llm_api_base is required (raises ValueError if empty)
• API key is optional for litellm (local proxy without auth)
• API key is forwarded when set (hosted proxy with auth)
• _validated_site_config() skips API key check for litellm
• _validated_site_config() still requires ai_model
• Other providers (openai) still require API key
• End-to-end get_llm_model() works for both local and hosted scenarios

Full test suite: 240 passed, 0 failures.

Manual testing against LiteLLM proxy (http://localhost:4000/v1) routing to Anthropic:

# Model discovery
curl http://localhost:4000/v1/models
# Returns: litellm/anthropic/claude-sonnet-4-20250514

# Completion through proxy
Provider: litellm
Model: anthropic/claude-sonnet-4-20250514
Result: successful completion with correct token counts

The builder follows the same AsyncOpenAI + OpenAIModel + OpenAIProvider pattern used by the existing openai_compatible provider, so streaming, tool calling, and structured output all work through the proxy.
](http://localhost:8000/admin/)

[RheagalFire]

@eracle

[RheagalFire]

@eracle do you have any update on this PR?