Fix compiler flag and further refactor autocorr loading.

In the last refactor I renamed some variables that broke compilation when running with the __DEBUG__ flag.

This should fix that, and I've added another minor refactor, just to starting to break down cuda_postcoh_load_autocorr_on_gpu into reusable chunks.

I've compiled with & without the debug flag, and it runs & prints ok. Haven't checked results yet, but given the kinds of changes, it's likely to be fine.