Enable build Intel backend in onemkl interfaces on CUDA (#2229)

This PR suggests to enable `MKLGPU_BACKEND` and `MKLCPU_BACKEND` builds
in OneMKL Interfaces during build on CUDA with `--target=cuda` flag to
ensure that all available devices can be used

Previous implementation only allowed array allocation on `cuda::gpu`
device with `ONEAPI_DEVICE_SEELCTOR=cuda:gpu` env variable enabled and
threw RuntimeError otherwise.

Author: vlad-perevezentsev

CMakeLists.txt

@@ -112,8 +112,8 @@ if(_use_onemkl_interfaces)
         set(ENABLE_CUSOLVER_BACKEND True)
         set(ENABLE_CUFFT_BACKEND True)
         # set(ENABLE_CURAND_BACKEND True)
-        set(ENABLE_MKLGPU_BACKEND False)
-        set(ENABLE_MKLCPU_BACKEND False)
+        set(ENABLE_MKLGPU_BACKEND True)
+        set(ENABLE_MKLCPU_BACKEND True)
     endif()
 
     if(DPNP_ONEMKL_INTERFACES_DIR)