Circular masks for intensity/tile features may be shifted by 1 pixel

[petebankhead]

Bug report

Describe the bug
When comparing intensity features computed with square and circular masks using Show measurement map, I noticed that there seemed to be a slight shift.

Digging into code, the relevant bit is starts here.

Adding a line

new ImagePlus("Something", new FloatProcessor(w, h, pixelImage.getArray(false))).show();

makes it possible to visualize the image in ImageJ. The screenshot suggests there could be an issue of shifting, at least if the width and height of the region is an even number.

If there is an odd number of pixels, the result looks ok:

(The contrast was changed differently in each case, to highlight the excluded pixels with NaN values)

Expected behavior
No apparent shift, regardless of whether there is an even or odd number of pixels in the image.

Desktop (please complete the following information):

• OS: All
• QuPath Version: Probably all up to v0.4.x

Additional context
This shouldn't be addressed in an 0.0.x release, since it would change results. But the circular tiles calculation should be revisited for a major release, since 1 pixel could be a substantial shift if the region is small (e.g. a low-resolution tile).