Astra-toolbox

* update Windows conda packages
* add support for python 3.12, cuda 12.5, and numpy 2.0

* astra conda packages for cuda 11.6 and above now depend on nvidia's
modular cuda packages instead of on the monolithic cudatoolkit package.
This improves compatibility of our conda packages with pytorch.
* speed up FDK, especially when using multiple GPUs
* improve error reporting

* fix set_gpu_index with Python 3.10
* fix set_gpu_index initializing GPU 0 before switching to the desired GPU

* fix memory leak in cone beam FP

* fix potential hang in CUDA FFT, and speed it up (for FBP_CUDA and FDK_CUDA)
* fix GPULink with rows not padded to multiple of 32 pixels
* fix output value scaling for short-scan FDK
* improve CUDA error logging
* fix compatibility issues with the conda-forge and nvidia conda channels

* fix inconsistent scaling of output value depending on detector pixel size
See the 2020-01-17 news entry on astra-toolbox.com for more details
* add experimental interfaces to FP3D, BP3D, FDK to make ASTRA calls
by 3rd party toolboxes more efficient and flexible
* improve compatibility with modern Python, CUDA, Linux and Windows versions