There are mainly following fixes:

- take `DeviceContext` as the template parameter of math functors and OpKernel instead of `Place`
- remove `eigen_device` interface in base class  `DeviceContext`
- remove `GetEigenDevice` interface in `ExecutionContext` and base class `DeviceContext`
- remove unused `platform::EigenDeviceConverter`
- rename `REGISTER_OP_GPU_KERNEL` to `REGISTER_OP_CUDA_KERNEL`
- rename `USE_GPU_ONLY_OP` to `USE_CUDA_ONLY_OP`

* Adding L1 norm op for L1 regularization

* Addressing code review feedback

* Address code review feedback

* Change variable names to match google style guide

* Adding the L2 loss operator for L2 regularization
* Renaming l2_loss op to squared_l2_norm_op
* Addressing code review feedback