Created by: NHZlX

1. add_transpose_flatten_concat_fuse In detect models like ssd, there exist patten of "transpose-> flatten ->concat ", we can fuse those op into one fused op to speed up inference time.

2. fix elementwise mul op converter bug.