Refactor device_malloc, queue::copy, and queue::fill calls

[kris-rowe]

Details

• Similar calls to sycl::device_malloc and sycl::queue::copy are repeated throughout sycl backends
  • Requires boilerplate code for retrieving a Ceed_Sycl pointer using CeedGetData
• These patterns can be refactored into wrapper functions which
  • Take a Ceed argument
  • Have an interface similar to cudaMalloc and cudaMemcpy
• Ensures all copies are synchronized in the same manner, e.g.
  • Currently queues are in-order and calls to queue::copy are followed by a wait_and_throw
  • Use of out-of-order queues will require either
    • Synchronization before calling queue::copy
    • Passing a vector of sycl::event dependencies