[PierreRaybaut]

Closes #101

Summary

Since #98, segment_profile sampled the segment in steps of the smallest distance between pixel centers (count = floor(length / step) + 1):

• the end point was omitted whenever the length was not a multiple of the step: the diagonal (0, 0)–(9, 9) of a 10 × 10 image gave 13 samples up to pixel (8, 8), duplicating some pixels and missing pixel (9, 9); a segment shorter than one step only sampled its start point;
• the number of samples did not depend on the number of pixels: a tiny spacing (non-uniform coordinates [0, 1e-8, 1, 2, 3], or dx = 1e-8) gave hundreds of millions of samples (several GB) for a few pixels.

Segment profiles are now sampled like in Sigima 1.3 (csline), in physical coordinates:

• max(|Δcolumn|, |Δrow|) + 1 samples, from the pixels nearest to the two end points;
• evenly spaced along the segment, end points included (exactly: weighted interpolation), each taking the nearest pixel; the X axis remains the physical distance from the start point;
• at most max(rows, columns) samples, whatever the pixel sizes.

The contract stays at version 2 (introduced by #98, not released yet). The release note no longer claims that all results are unchanged on images without calibration: horizontal, vertical and diagonal segments take the same pixels as in Sigima 1.3, other oblique segments take the nearest pixel instead of truncating its indices (French translation updated, checked with a gettext build + sphinx_intl update).

Companion: DataLab-Platform/DataLab#440 (same branch name) updates the description of the segment profile in the DataLab documentation.

Validation

• New tests (profile_physical_unit_test.py): end points included on diagonal and oblique segments, identical to Sigima 1.3 for axis-aligned and diagonal segments (6 cases, reference reimplemented from 1.3), segments shorter than a pixel / within one pixel / of zero length, tiny spacings (non-uniform and anisotropic: 5 samples for a 5 × 5 image).
• Validation test updated deliberately: the anti-diagonal segment now gives 31 samples (one per pixel, end pixel included) instead of 43.
• Full suite: 2345 passed; --qualification-report: no problem (segment_profile v2 qualified, validation and replay passed).