Support automatically offloading of NumPy reduction operations and prange reductions to SYCL devices

[diptorupd]

Numpy reduction operations are not yet automatically offloadable to a SYCL GPU. To get around the issue, we are detouring to fallback on CPU execution for all reduction functions (refer PR IntelPython/numba#8). We need to support reductions by either directly generating a SPIR-V kernel for reduction inside Numba or calling a library routine that does the reduction.

[diptorupd]

@reazulhoque The list of functions that are planned is needed for this ticket so that we can implement and close the ticket.

[diptorupd]

Currently, NumPy reduction calls such as sum, product, dot are supported via dpnp library calls. Closing the issue as it is too broad, we can open newer issues for specific routines as needed.