Measure the slice extent along the slice axis, not along patient z

[gdevenyi]

Fixes #194.

"cmd": "np.array([#VisuCoreExtent[0], #VisuCoreExtent[1], (np.shape(#VisuCorePosition)[0] + 1 ) * abs(#VisuCorePosition[1,2] - #VisuCorePosition[0,2])])"

Two errors in one expression: only the z component of the slice step is used, and the step is counted N + 1 times.

Spec 7.2 makes VisuCorePosition a 3-vector in patient coordinates; spec 5.4/12 let the slice normal point anywhere (ACQ_slice_orient ranges over sagittal, coronal and Arbitrary_Oblique, and the normal is the third row of VisuCoreOrientation). Projecting onto z alone is zero whenever the stack is not axial.

Before / after

.../11/pdata/1  step [0,-1,0]  10 slices x 1mm   extent [20, 20,  0.0] -> [20, 20, 10.0]
.../6/pdata/1   step [0, 0,1]  10 slices x 1mm   extent [20, 20, 11.0] -> [20, 20, 10.0]

Corpus sweep over pv5, pv6, pv7, new-orientation:

before: 77 multi-slice 2-D reconstructions, 14 with extent[2] == 0, 63 off by one slice
after : 77 correct, 0 wrong

Change

np.shape(#VisuCorePosition)[0] * np.linalg.norm(#VisuCorePosition[1,:] - #VisuCorePosition[0,:]) — the length of the whole difference, which is what the sibling resolution recipe and Dataset.slice_distance already use, times the number of slices, which keeps extent = size × resolution as it is in plane.

The corpus property references move with it: 54 extent values across the four test/config/properties_*.json, regenerated in place so key order and layout are unchanged.

Test

test_extent_spans_the_slice_axis_whatever_direction_it_points builds a synthetic 5-slice stack for an axial, coronal, sagittal and oblique normal and asserts extent[2] == n × step, agreeing with resolution[2] and slice_distance. Three of the four parameters fail on master.

Suite: 2140 passed, 12 skipped.

[gdevenyi]

Rebased onto #213 (ruff format), so this branch carries that commit until #213 merges — merge #213 first and this diff reduces to its own change. Verified after the rebase: ruff check and ruff format --check clean, full suite green.

Integration check across all eleven conformance PRs merged together: 2164 passed, 12 skipped; corpus load test 3202 → 3207 of 3478 with zero newly broken datasets.