…er ordering

Three related bugs fixed in the slab material layer set workflow:

1. EditAssignedMaterial (operator.py): Applying a custom offset to one slab
   instance incorrectly regenerated geometry for ALL slabs sharing the same
   IfcMaterialLayerSet. Replaced regenerate_from_layer_set (sweeps all users)
   with per-element regenerate_from_occurence for AXIS3 slabs and targeted
   recalculate_walls for AXIS2 walls. Each element's IfcMaterialLayerSetUsage
   attributes are now updated individually before regeneration.

2. slice_layerset_mesh (loader.py): Loader.unit_scale is a class variable only
   set during full file import, so it was stale (= 1) during live geometry
   updates on foot-based IFC files. Layer bisect planes were being computed in
   IFC feet while the mesh was in Blender metres, placing all cuts completely
   outside the mesh. Fixed by computing unit_scale fresh from the IFC file on
   each call via ifcopenshell.util.unit.calculate_unit_scale.

3. OffsetFromReferenceLine stale / layer order reversed (slab.py, loader.py):
   A guard (and custom_offset is None) in change_thickness prevented writing
   the correct OffsetFromReferenceLine (position.z) to the usage when a custom
   offset was active. This left the value at 0.0 instead of the actual slab
   bottom (e.g. -1.0 IFC units for a TOP-reference slab), so the bisect
   starting point co was at the reference plane rather than the slab bottom,
   reversing layer assignments or missing layers entirely. Removed the guard —
   safe because each element has its own IfcMaterialLayerSetUsage instance.
   Reverted the AXIS3 bisect normal back to (0,0,1) (upward from co at slab
   bottom) which is correct once OffsetFromReferenceLine is properly set.