Feature Request: Add MKL-DNN to speed up dynet conv2d on CPU

Dynet is a great deep learning framework. However, it's conv2d implementation is very slow on CPU, about 6 times slower than PyTorch on a 6700K-CPU machine. Recently Intel released the [MKL-DNN](https://github.com/intel/mkl-dnn) library which can significantly improve Conv2d, Relu, BatchNorm etc performance on Intel CPU platform. For those of us who don`t have GPU, it will be a good News to have it in Dynet.

The issue identifies Dynet's CPU conv2d implementation and MKL-DNN as the starting points; review MKL-DNN support for Conv2d, Relu, and BatchNorm. Compare CPU conv2d with the current implementation on the stated 6700K setup, but note that the issue names no files or tests and does not define a completion criterion.

Written by the indexing model from the issue text.