Fix build (part1):

1. Changed includes of "NvInfer.h" to "tensorrt/include/NvInfer.h"
2. Remove build target "tensorrt_ops.so" (src file doesn't exist and the
   target is not used anywhere)
3. Add missing '#if GOOGLE_TENSORRT's
4. Use tf_cuda_library instead of cc_library for some targets to get the
   tf_copts naturally.
5. Revert the changes that was accidentally made (by merging with
   upstream head) from configure.py
6. Replace exception with LOG(FATAL) in
   tensorflow/contrib/tensorrt/convert/convert_nodes.cc as exception is
   not supported.
7. Revert the reinterprete_cast change in
   tensorflow/contrib/tensorrt/kernels/trt_engine_op.cc
8. Fix minor formatting and naming issues according to the style guide.