[DFT][MKLGPU] Use FWD_ and BWD_STRIDES for MKLGPU 2024.1

[hjabird]

Summary

The MKLGPU library should use updated oneMKL interface for forward and bwd strides for the MKLGPU backend.

Problem statement

The oneMKL specification was revised to use FWD_ and BWD_STRIDES instead of INPUT_ and OUTPUT_STRIDES. This has been implemented in Intel's oneMKL for 2024.1. Whilst oneMKL interface library still works, MKLGPU prints deprecation notices. See https://www.intel.com/content/www/us/en/developer/articles/release-notes/onemkl-release-notes.html.

Preferred solution

When using Intel oneMKL 2024.1 or later, use the updated API for strides.