Use get_acc_offload() where appropriate

[nmnobre]

Made possible by #196. Does not change semantics, simply reuses code for robustness/conciseness.

[xiaoyeli]

@nmnobre Does it solve this problem? - even if you compile the code with GPU enabled (such that GPU_ACC is defined), but you don't want to use GPU offload (i.e., setting superlu_acc_offload=0), then there should be no GPU memory allocated?

[nmnobre]

even if you compile the code with GPU enabled (such that GPU_ACC is defined), but you don't want to use GPU offload (i.e., setting superlu_acc_offload=0), then there should be no GPU memory allocated?

This was already the case. It's my bad for not realising that offloading was on by default. This PR (and now #207) simply refactored the code for code reuse/readability/conciseness, it does not change the code behaviour whatsoever.