GHSA-Q2C3-JPMC-GFJX TensorFlow vulnerable to `CHECK` fail in `Conv2DBackpropInput`

Impact The implementation of Conv2DBackpropInput requires inputsizes to be 4-dimensional. Otherwise, it gives a CHECK failure which can be used to trigger a denial of service attack: python import tensorflow as tf strides = 1, 1, 1, 1 padding = "SAME" usecudnnongpu = True explicitpaddings =...