* one euro filter and ema smoothing for keypoints

* change few code
Co-authored-by: NZhangHandi <handi.zhang@163.com>

* add keypoint benchmark data

* update benchmark-data and change default keypoint bs=8, to be same as C++ infer

* support rcnn bs=2 in cpp infer

* check dynamic shape in padbatch

* replace fluid by paddle

* fix time display in pptracking cpp

* fix timer in bs=2

* merge upstream

* clean code

* add batch_size in keypoint_detector

* refine keypoint deploy

* fit video infer

* fix post process

* update comments for keypoint_batch_size

add dark deploy supported, mpii deploy supported;

* support xpu inference

* Automatically set use_dynamic_shape