CCPi Regularisation Toolkit Versions Save

• replace cython with ctypes (#183 <- #178, #177, #164)
• fix conda build (#189 <- #187, #192)
  • fix python path
  • native cmake --build & pip install
  • build into build_proj subdir
  • fix package name (meta.yaml vs pyproject.toml consistency)
  • hardcode version in pyproject.toml (#185) as 24.0.0
  • fix windows builds
• migrate from @vais-ral to @TomographicImaging (#181)
  • update documentation links (#202)
• use GitHub Actions CI (#193)
  • self-hosted CUDA runners (#196 <- #195)
    • force cleanup on setup fail (#201)
  • test windows & more versions (#203 <- #195)
• bump python, numpy (#200 <- #194)
• add cupy optional dependency (#199)

MultiGPU capability to control devices has been added and the following demo shows how to run it with mpi4py package. Now regularisation_device can be passed as 'cpu' or 'gpu' OR provide a GPU index (integer) of a specific device

• nested loops re-organised which gives a substantial acceleration running CPU modules
• NLTV gpu bugs fixed