feat: integrate LazyLLM to provide more llm services

[uestcsh917-art]

Feature Overview

• Add LazyLLM backend to support to extend MemU's multi-model AI service capabilities, compatible with LLM/VLM/Embed/STT model types.
• Provide 1 complete example programexamples/example_5_with_lazyllm_client.py covering conversation memory processing, skill extraction, and multimodal memory generation scenarios.
• Provide 1 test file tests/test_lazyllm.pyto verify LazyLLM backend configuration and basic functionality

Core Changes

1. New LazyLLM Client Implementation (src/memu/llm/lazyllm_client.py)

• Implemented LazyLLMClient class with a unified interface.
• Support 4 model capabilities and multiple model providers:
  • Source(model service provider): Default qwen
  • LLM (Text Generation/Summarization): Default qwen-plus, provides summarize() method
  • VLM (Vision Language Understanding): Default qwen-vl-plus, provides vision() method
  • Embed (Text Embedding): Default text-embedding-v3, provides embed() method
  • STT (Speech-to-Text): Default qwen-audio-turbo, provides transcribe() method
• Key Features:
  • Asynchronous call support.
  • flexible providers/models configuration.

2. One New Example Programs

Example File	Functional Scenario	Output Path
examples/example_5_with_lazyllm_client.py	1. Multi-conversation file processing → Memory category generation 2. Workflow/agent logs → Skill extraction (task guidelines) 3. Document + image multimodal processing → Unified memory categories	examples/output/lazyllm_example/

Testing Steps

Prerequisites

• Install lazyllm dependencies: pip install lazyllm
• Configure API key according to the source and model configuration in tests/test_lazyllm.py,Defaut qwen
export MEMU_QWEN_API_KEY='your-api-key'

Test Commands

# Run unit tests
python tests/test_lazyllm.py

# Run example programs (verify end-to-end functionality)
python examples/example_5_with_lazyllm_client.py