we can improve the performance for xgboost transformation part on GPU

[nvliyuan]

we can improve the performance for xgboost transformation part on GPU,
I test mortgage xgboost notebooks in the internal cluster and find the transformation part is much slower(8 times) than CPU, will update the xgboost jars to 1.7.0rc1 and update the codes to verify the lastest result, just file an issue to track

[YuanLinNV]

@nvliyuan , maybe you want to assign this ticket to someone else?