feat: add configurable batch_size for embedding API calls

[zhixiangxue]

Problem

Different embedding API providers have different batch size limits:

• OpenAI: ~25 items per batch (no strict limit documented)
• Bailian/DashScope: max 10 items per batch (strict limit)
• Other providers may have different limits

Currently, MemU processes all embeddings in a single batch, which causes errors when the batch size exceeds the provider's limit.

Solution

This PR adds a configurable batch_size parameter to handle provider-specific limits:

1. Added batch_size to EmbeddingConfig

   • Default: 25 (suitable for OpenAI)
   • Users can configure it based on their provider (e.g., 10 for Bailian)

2. Implemented batch processing in OpenAIEmbeddingSDKClient

   • Automatically splits large input lists into smaller batches
   • Optimized: skips batching when input size <= batch_size

3. Updated service initialization

   • Passes batch_size from config to embedding client

Changes

• src/memu/app/settings.py: Add batch_size field to EmbeddingConfig
• src/memu/embedding/openai_sdk.py:
  • Add batch_size parameter to __init__
  • Implement batch processing in embed() method
• src/memu/app/service.py: Pass batch_size to embedding client

Example Usage

from memu.app import MemoryService

# For Bailian/DashScope (max 10 per batch)
embedding_config = {
    "base_url": "https://dashscope.aliyuncs.com/compatible-mode/v1",
    "api_key": "YOUR_KEY",
    "embed_model": "text-embedding-v3",
    "batch_size": 10  # Configure batch size
}

service = MemoryService(embedding_config=embedding_config)