Multisem inpainting

[minnerbe]

This PR updates the inpainting code for wafer 60/61 data:

• For every sfov in the stage coordinates of an aligned stack, it computes how far that sfov is from the ROI as given in the xlog data.
• The first layer of the aligned stack is treated as the true position of sfovs and a point cloud is generated from that (coordinates = sfov centers from render, values = distance from ROI from xlog).
• For the exported stack (N5), each chunk is put in the global coordinate system its distance to the ROI is computed by interpolation from the point cloud.
• For all chunks that are close enough to the ROI (i.e., closer than a given threshold in micron), pixels are inpainted if the mask has a values < 255.
• If a chunk is modified, it is saved in the original dataset and the original chunk is backed up to a different dataset (so we could go back).

This seems to work reasonably well: it's fairly quick, requires no manual intervention, and even caught some holes that were way too small for me to see on a downsampled mask. From a method point-of-view, I think this is done.

@trautmane let me know if I should add some support for multi-stack setups.