mumax 3.9.1

This version works around issue #52: a bug in NVIDIA's cuFFT library caused wrong FFT results for certain sizes with CUDA >= 6.5. Fortunately, a sanity check in mumax detected this issue, causing mumax to checkfail and not return wrong results. Thanks @godsic for detecting and fixing this issue.

Upgrading is recommended for everyone using CUDA 6.5 or newer.