LIF: fix offset calculation for certain datasets with multiple unstitched files

[melissalinkert]

Fixes #3873, fixes #4199.

See files in inbox/qa/31243 and inbox/gh-3873. Without these changes, showinf should throw an ArrayIndexOutOfBoundsException as described in the issues. With these changes, both files should initialize and display images without exception.

ImageJ may be the easiest way to test, as much of this data has small pixel values, so the images may be hard to check with just showinf. Composite color mode may also be useful to more closely match what LAS X shows. Screenshots of LAS X for comparison will be added to the forthcoming config PR.

I'd expect this to be OK for a patch release assuming tests continue to pass.

[sbesson]

Tested using the sample files uploaded as part of the corresponding issues.

Without this PR included, the reader initialization fails with an IndexOutOfBoundsException as described and any downstream workflows like importing into OMERO are broken.

With this PR included, the reader initialises without issue for both sample files and the images reading is consistent with the ground truth as shown in the LAS X software.

The nightly CI builds have been passing successfully for the last days with this PR included showing there are no apparent regression for other LIF sample files.