The locally built 0.9 runtime (all 59 DLLs incl. the 8 geometry-mapping
plugins missing from official win64 artifacts, upstream IfcOpenShell#9301/IfcOpenShell#9305)
un-gated 84 geometry-dependent tests. Running them surfaced:

- _add_segment_to_curve: only refresh EndPoint via update_end_point when
  the gradient/segmented-reference curve already carries its zero-length
  terminal segment; mid-construction flows (create_representation) add
  segments before the terminal exists and the fallback path cannot
  resolve transition codes against a curve still being assembled.
- segment_vertices (+ API test): entity_instance no longer exposes
  .wrapped_data in the 0.9 wrapper; pass instances directly to map_shape.
- _map_linear_transition: guard start_direction against A1 == 0 — a
  LINEARTRANSITION cant segment with equal start/end cant is schema-legal
  and the expression's limit as the cant change goes to 0 is 0 (was
  ZeroDivisionError).
- CIVIL_OT_add_cant_point: move the display-row selection to the point
  just added; insert-after resolution reads active_cant_display_row_index,
  so leaving it behind made every subsequent Add insert midpoints after
  the original point instead of appending.
- Operator-test fixtures: create_alignment_with_horizontal_and_vertical
  now lays out a real 1000 m tangent (cant validation correctly refused
  the old zero-extent bare alignment); equal back/ahead station-equation
  test asserts the RuntimeError bpy.ops raises in background mode.

Bonsai alignment suites: 503 passed / 0 failed / 0 skipped (was 499/4,
with 84 tests formerly skip-gated). API alignment: 110 passed. Core: 121.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

…ment_vertices helpers

Fixes found while building Bonsai alignment authoring on top of the
alignment API:

- _create_geometric_representation: typo creatIfcShapeRepresentation ->
  createIfcShapeRepresentation (crashed 3D representation of gradient
  curves).
- get_mapped_segments: off-by-one — the second mapped segment of a
  two-segment mapping returned Segments[index] instead of
  Segments[index - 1].
- get_stationing_nest: matched the first nest containing any IfcReferent;
  now requires PredefinedType == "STATION" so position referents don't
  shadow the stationing nest.
- create_as_offset_curve: offset-curve alignments never received a
  stationing referent; now establishes the stationing scheme exactly as
  create() does for start_station.
- _get_segment_start_point_label: vertical CONSTANTGRADIENT/
  CONSTANTGRADIENT label "P.V.I" -> "P.V.I.".
- segment_vertices: restored helper returning (Start, End, TI, NI)
  control points per segment — needed by any PI-based editor and gone
  from the current API surface.

New Saikei helper: clear_layout_segments — the layout functions only
append and expose no clear/remove, so editing a layout (PI recalc, edit
mode) needs a safe way to drop real segments while keeping the
zero-length terminator and cleaning up mapped geometry without orphans.

Tests included for all of the above (test/api/alignment).
- segment_vertices: pass the entity_instance to map_shape directly —
  entity_instance.wrapped_data is gone in the 0.9 modular wrapper
  (verified against a locally built 0.9 runtime with the per-schema
  geometry-mapping plugins that the official win64 artifacts omit,
  upstream IfcOpenShell#9301/IfcOpenShell#9305).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

The locally built 0.9 runtime (all 59 DLLs incl. the 8 geometry-mapping
plugins missing from official win64 artifacts, upstream IfcOpenShell#9301/IfcOpenShell#9305)
un-gated 84 geometry-dependent tests. Running them surfaced:

- _add_segment_to_curve: only refresh EndPoint via update_end_point when
  the gradient/segmented-reference curve already carries its zero-length
  terminal segment; mid-construction flows (create_representation) add
  segments before the terminal exists and the fallback path cannot
  resolve transition codes against a curve still being assembled.
- segment_vertices (+ API test): entity_instance no longer exposes
  .wrapped_data in the 0.9 wrapper; pass instances directly to map_shape.
- _map_linear_transition: guard start_direction against A1 == 0 — a
  LINEARTRANSITION cant segment with equal start/end cant is schema-legal
  and the expression's limit as the cant change goes to 0 is 0 (was
  ZeroDivisionError).
- CIVIL_OT_add_cant_point: move the display-row selection to the point
  just added; insert-after resolution reads active_cant_display_row_index,
  so leaving it behind made every subsequent Add insert midpoints after
  the original point instead of appending.
- Operator-test fixtures: create_alignment_with_horizontal_and_vertical
  now lays out a real 1000 m tangent (cant validation correctly refused
  the old zero-extent bare alignment); equal back/ahead station-equation
  test asserts the RuntimeError bpy.ops raises in background mode.

Bonsai alignment suites: 503 passed / 0 failed / 0 skipped (was 499/4,
with 84 tests formerly skip-gated). API alignment: 110 passed. Core: 121.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

The locally built 0.9 runtime supplies the eight geometry-mapping plugins
missing from official win64 artifacts (IfcOpenShell#9301/IfcOpenShell#9305), so all suites run
with zero skips. Records the build inputs kept for rebuilds, the runtime
backup path, and the TODO to swap back to official artifacts once IfcOpenShell#9305
merges (bit-parity with what reviewers and CI run).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>