[model loading] Add max_batch_size override for batch size exploration
Summary: Currently batch_size is determined on modeling side. Add a flag caffe2_predictor_disagg_acc_max_batch_size_override to explore different batch_size during inference.
Test Plan:
replayer test
set caffe2_predictor_disagg_acc_max_batch_size_override=32 on both server and client side.
Reviewed By: khabinov
Differential Revision: D26318568
fbshipit-source-id: 4fa79e2087a5f7f7670988aec7e5b41e63f9980b