Double flatfield: issues when using LimitedMemory pipeline

Reported by Ludo and Bratislav.

In some cases, the "double flatfield" works differently when used on a single slice, on a whole volume, or on a subvolume.

The likely reason is that the underlying pipeline CudaFullFieldPipelineLimitedMemory computes one DFF for projections [0, ..., n_1], then another DFF for projections [n_1 + 1, ..., n_2], and so on. The "true" DFF is different than DFF computed on sub-stack of projections.

There is no solution for CudaFullFieldPipelineLimitedMemory and FullRadiosPipeline outside computing the DFF in a first stage. Therefore, a CLI tool should be provided to generate a DFF file, which can be used afterwise.