Updates for Volta

[sharannarang]

• Add support for Tensor Cores to GEMM, RNN & Convolution benchmarks. This feature requires the use of CUDA 9.0 and cudNN 7.0 or higher.
• Add V100 results file containing results for single precision, single precision inputs with tensor core math and half precision inputs with tensor core math.
• Update README with Volta results and instructions to build with Tensor Core support.

Note: AllReduce results aren't available for Volta as yet. They will be available in the next couple of weeks.

[sharannarang]

This PR has been reviewed internally by researchers at Baidu. Merging this PR.