横向联邦中线性模型的训练和测试代码统一性修复

[imxuebi]

fix: unify HomoLR predict logic and handle batch_size

Changes:

1. Refactor HomoLRClient.predict in python/fate/ml/glm/homo/lr/client.py:

   • Always initialize a new FedAVGClient instance with local_mode=True for prediction.
   • This ensures consistent prediction behavior regardless of whether it is called during training (validation) or as a standalone prediction task, avoiding side effects from the training trainer state.

2. Update predict component in python/fate/components/components/homo_lr.py:

   • Add handling for batch_size <= 0.
   • If batch_size is non-positive, it is now correctly converted to None (indicating full batch), preventing errors when initializing HomoLRClient.