Add triSYCL support for Tensorflow

Here is a description of the changes brought to each file :

configure :

- Add the option to choose triSYCL when choosing to configure Tensorflow
  with OpenCL, this will require the user to give the path to the
  triSYCL include directory

tensorflow/core/common_runtime/sycl/sycl_device.h :

- Add the SYCL host device as a device that can be used for
  computations, currently triSYCL only supports the host device
  throught OpenMP parallelism

tensorflow/core/kernels/ransom_op.cc :

- Fix calls to item.get_global_range that should not take any arguments
  as specified in the SYCL specification

tensorflow/core/platfom/default/build_config/BUILD :

- Modify the BUILD file to not include ComputeCPP dependencies when
  using triSYCL

tensorflow/stream_executor/platform/default/mutex.h :

- Modify mutex to use std::condition_variable_any, std::shared_lock
  and std::shared_mutex when using C++14

tensorflow/workspace.bzl :

- Update the version of Eigen used to a version compatible with triSYCL

third_party/sycl/crosstool/CROSSTOOL.tpl :

- Made the config file able to take triSYCL variable to configure crosstool

third_party/sycl/crosstool/tristcl.tpl :

- Create the triSYCL compiling script file

third_party/sycl/sycl/* :

- Modify the configuration files to enable triSYCL

tools/bazel.rc :

- Add the --opt=sycl_trisycl option to enable triSYCL with Tensorflow