Make displayed and actual point analysis coordinates match

[uellue]

How to reproduce

• LiberTEM current master, possibly older versions as well
• Open a low-resolution dataset, for example 32x32x32x32, with clearly identifiable points (in this case a double diagonal with outliers).
• Perform a point analysis around the outliers. Disk analysis and frame picking are not affected, others not tested.

Expected results

• A pixel on the left is selected if the center of the crosshair is within it.
• As a bonus, the selector position could snap to pixel centers like it does for frame picking.

Observed result

• The selection area for a pixel seems to be shifted by 1/2 to the lower right. Based on the numerical values, one should use "round()" instead of "floor()" to get the integer point position from float coordinates.

See these pictures:

Centered: correct pixel

Lower left corner: Wrong pixel selected

Lower right corner: Correct pixel selected

Upper right corner: Wrong pixel selected

Upper left corner: Pixel to the upper left selected (wrong pixel, other outlier)

Just below pixel: Still within selection

Just right of pixel: Still selected

[sk1p]

Thanks for the detailed report, this should be fixed in #927