Speed up dcf calculation

[ckolbPTB]

Currently the dcf calculation is very slow, specifically this part: https://github.com/PTB-MR/mrpro/blob/025bc796b845aabdf255b2ed584f0ff26ae0cb28/src/mrpro/data/_DcfData.py#L118C18-L118C36

I tried to exchange the ThreadPoolExecutor with multiprocessing.Pool but this did not make any change.

Possible ways to solve this: Find a pytorch compatible way of calculation Voronoi cells or at least replace ConvexHull(v).volume by a function running on GPU

[fzimmermann89]

The issue is that the number of vertices in each cell is different. This makes vectorization difficult.
I will keep this as a todo until the next long train ride and would test if we can write a small separate utility that also wraps qhull with cython, but allows a list of arrays of vertices and internally parallelizes over the list.

[fzimmermann89]

For 3D there is either https://github.com/eleftherioszisis/tess which unfortunately does not have CD setup (there is an unmerged uncommented PR implementing compilation and pushing to pypi) or maybe something will come out of this request at scipy scipy/scipy#20118

[fzimmermann89]

Scratch that. At least the "Tessa" python wrapper is way too inefficient. Maybe directly inferacing voro could be fast enough..

[fzimmermann89]

scipy/scipy#20118

and #629