Migrate the raytracing benchmark

[jinz2014]

"the resulting SYCL version of Raytracing differs significantly from the original CUDA, and hence, their execution times are not directly comparable. The two main reasons for that are 1) the different random number generator introduced by DPCT, namely oneMKL’s philox4x32x10 [15], vs. the original cuRAND’s XORWOW [21]), and 2) the significant manual code refactoring we required to cope with CUDA virtual functions"

The paper is attached.
3624062.3624542.pdf

[tomflinda]

@jinz2014 thanks for your report, we will do a further investigation.