feat: add LiteLLM as AI gateway provider

[RheagalFire]

Summary

• Adds LiteLLM as a native provider alongside Ollama, OpenAI, Gemini, OpenRouter, Mistral, DeepSeek, and Poe
• Enables access to 100+ LLM providers via provider-prefixed model names (e.g. anthropic/claude-sonnet-4-6, bedrock/anthropic.claude-v2)

Motivation

TranslateBooksWithLLMs already supports 7 providers through individual implementations. LiteLLM adds a unified gateway that handles provider-specific API differences, parameter translation, and authentication —
letting users access any of 100+ providers without needing a separate adapter implementation for each.

Changes

• src/core/llm/providers/litellm.py — New LiteLLMProvider extending LLMProvider ABC with generate(), retry logic with transient error detection, and context overflow handling
• src/core/llm/factory.py — Added "litellm" case to create_llm_provider() factory
• requirements.txt — Added commented optional litellm>=1.65,<1.85 dependency
• tests/unit/test_litellm_provider.py — 4 unit tests covering completion dispatch, credential forwarding, system prompt handling, and factory integration

Tests

1. Unit tests: pytest tests/unit/test_litellm_provider.py -v
tests/unit/test_litellm_provider.py::test_generate_calls_acompletion PASSED
tests/unit/test_litellm_provider.py::test_generate_omits_blank_credentials PASSED
tests/unit/test_litellm_provider.py::test_generate_forwards_system_prompt PASSED
tests/unit/test_litellm_provider.py::test_factory_creates_litellm_provider PASSED
4 passed in 0.12s

Risk / Compatibility

• Additive only — existing providers untouched
• litellm is optional — not added to core requirements
• LiteLLM lazy-imported inside generate() to avoid import errors when not installed
• drop_params=True by default for cross-provider kwarg compatibility

Example usage

from src.core.llm.factory import create_llm_provider
                                                                                                                                                                                                                   
provider = create_llm_provider("litellm", model="anthropic/claude-sonnet-4-6", api_key="sk-...")
response = await provider.generate("Translate: Hello world", system_prompt="Translate to French")                                                                                                                  
print(response.content)                                                                                                                                                                                            

Or via CLI: set LLM_PROVIDER=litellm and MODEL=anthropic/claude-sonnet-4-6 in your .env.

[RheagalFire]

cc @hydropix