support CUDA 13.1, SM 12.0 by ktdq · Pull Request #3696 · arrayfire/arrayfire

Fix #3695

support more cuda arch — expands CUDA architecture handling (adds newer SM targets).
Handle optional cuFFT enums and use queried clock rate — improves CUDA backend robustness with newer/cuFFT changes.

Vibe coded with OpenAI codex, probably all wrong, but works.