Fix lstm possible leak

test=release/1.3

Fix security issue

test=release1.3

remove legacy V2 code in release1.3

test=release/1.3

Fix old FC backward weights descriptor creation

Cherry-pick debug mode fix to release/1.3

test=release/1.3

fix generate doc error in activation ops

test=release/1.3

fix the build issue on gpu mode for win

1.3/fixci

test=release1.3

test=release1.3

test=develop

Ir inplace pass

optimize density_prior_box_op.h in detect model for cpu

test=develop, fast_install shell for linux and mac

* update_readme

* test=develop

To make CUDA_LAUNCH_KERNEL_HELPER support large size.

test=develop