Re-implement SYCL backend `parallel_for` to improve bandwidth utilization #535

Job	Run time
HOST,bknd=tbb,cmplr=icx-cl,windows-latest,std=c++17,cfg=release	19m 37s
HOST,bknd=tbb,cmplr=cl,windows-latest,std=c++17,cfg=release	16m 50s
clang-format	17s
CPU,bknd=dpcpp,cmplr=icx,windows-latest,std=c++17,cfg=release	10m 25s
CPU,bknd=dpcpp,cmplr=icpx,ubuntu-latest,std=с++17,cfg=release	15m 30s
HOST,bknd=omp,cmplr=clang++,macos-latest,std=c++17,cfg=release	4m 44s
FPGA_EMU,bknd=dpcpp,cmplr=icpx,ubuntu-latest,std=с++17,cfg=release	24m 9s
HOST,bknd=tbb,cmplr=icpx,ubuntu-latest,std=с++17,cfg=release	12m 21s
HOST,bknd=omp,cmplr=icpx,ubuntu-latest,std=с++17,cfg=release	9m 54s
HOST,bknd=omp,cmplr=g++,ubuntu-latest,std=с++17,cfg=release	6m 30s
HOST,bknd=tbb,cmplr=g++,ubuntu-latest,std=с++17,cfg=release	9m 52s
HOST,bknd=tbb,cmplr=g++,ubuntu-latest,std=с++17,cfg=release	9m 55s
HOST,bknd=tbb,cmplr=clang++,ubuntu-latest,std=с++17,cfg=release	11m 18s
HOST,bknd=omp,cmplr=clang++,ubuntu-latest,std=с++17,cfg=release	8m 21s
HOST,bknd=tbb,cmplr=clang++,ubuntu-latest,std=с++17,cfg=release	11m 17s
HOST,bknd=tbb,cmplr=icpx,ubuntu-latest,std=с++17,cfg=release	12m 12s
HOST,bknd=serial,cmplr=g++,ubuntu-latest,std=с++17,cfg=release	5m 17s
CPU,bknd=dpcpp,cmplr=icpx,ubuntu-latest,std=с++20,cfg=release	26m 20s
HOST,bknd=tbb,cmplr=icpx,ubuntu-latest,std=с++20,cfg=release	17m 36s
HOST,bknd=omp,cmplr=icpx,ubuntu-latest,std=с++20,cfg=release	18m 9s
	4h 10m 34s

Provide feedback