[lachlangrose]

Step 1 of consolidating the structured-grid geometry code. No behaviour change for the interpolation path.

Background

StructuredGrid3DGeometry is a copy of the geometry half of BaseStructuredSupport (24 of its ~25 members are shared). It has drifted. The only in-library user is the aabb_grid of the UnstructuredMesh*Geometry classes, and it had no tests. Before it can be used as the single geometry type (a later step), it has to agree with the support. This PR makes it agree and pins that with tests.

What was different

Found by running both classes on the same inputs (unit cube, offset anisotropic box, 30-degree rotated box):

Member	Geometry class (before)	Support (and now geometry)
global_node_indices	stub, returned its input, so elements was (n, 8, 3) ijk triples	flat node indexes, elements is (n, 8)
position_to_cell_index	truncating, no clip, no outside handling: x=6 on a 4-cell grid gave cell 6	floor, clipped to the last cell, outside points get (0,0,0) and inside=False
inside	strict > / <, boundary excluded, no tolerance, no input validation	boundary included to a small tolerance, validates input
global_index_to_node_index	1-D only	any shape
global_cell_indices	missing	added

global_index_to_cell_index also dropped a modulo on the k index. It gives the same result for valid indices.

Tests

packages/loop_common/tests/test_structured_grid_geometry.py (22 tests) compares the geometry class with loop_common.supports.StructuredGrid on three grids, and covers the maximum boundary, outside points, single points and bad shapes. 19 of the 22 fail on the old code; the other 3 were already correct.

• packages/loop_common/tests: 178 passed
• packages/loop_interpolation/tests: 606 passed, 46 skipped, 2 xfailed
• tests: 730 passed, 6 skipped
• ruff: no new findings

Notes

• Behaviour change: inside() and position_to_cell_index on StructuredGrid3DGeometry now include boundary points. I found no callers outside the library besides the aabb_grid use, but I have not checked the QGIS plugin or map2loop.
• Neither this class nor the support applies rotation_xy in position_to_cell_index or inside; points must already be in the grid's own frame. The docstring now says so.
• Not in this PR: the 2D class (same pattern, same drift likely), the other support members (position_to_cell_corners, node_indexes_to_position, element_size, vtk), and making BaseStructuredSupport extend this class.

🤖 Generated with Claude Code