Add Intel FPGA code paths to SYCL back-end

[j-stephan]

This is basically the Intel variant of #1602 (with the exception of parallel_for not being broken). For FPGAs we should add a single_task kernel and make use of Intel's SYCL extensions where possible. Intel provides special load/store extensions; maybe we can use these after we have found a solution to #38 and #1312.