Accurately build half-acquisition sinogram

For now, half-acquisition final sinogram is built in a very simple way: two half-sinograms are stitched together, with a linear weighting in the overlap region.
This approach has two drawbacks:

• It does not work for odd number of projections (which unfortunately is usually the case in half-tomo scans at ESRF)
• It assumes that each pair of projections are separated by exactly 180 degrees - which is also usually slightly wrong and gives "blur" artefacts.

A correct sinogram builder has to be implemented, probably with bilinear interpolation.