Created by: juncaipeng

• Collect output scale for quantized op and fused op in post_training_quantization
• Add api for setting activation_quantize_type and weight_quantize_type
• Add api for saving combined model
• Post_training_quantizaion sets batch_generator to support lod tensor