orgin map thrust transform to direact kernel call

storage with kernel map

pytorch 1.8.1 support with gpu_storage

transpose kernel map with stride==kernel fix

fix reconstruction example (fix #323)

dispatcher cleanup, pytorch requirement update

conda cudatoolkit update

Readme update for unstable pytorch 1.8

migration guide in doc

update segmentation image, indoor example minor update

separate memory chunks for gpu kernel map

global max fix for batchsize 1

TensorField batch row index using hashtable

fix broken links

updated training augmentation in modelnet40