refactor: split interactive_structure_editor.ts into focused mixins (TB-ARCH-2)

[timurbazhirov]

Summary

• interactive_structure_editor.ts was 1460 lines, flagged by the TB-persona review (TB-ARCH-2) as "still the largest file in src/mixins/ by a wide margin." Extracts the two most cleanly-separable feature areas into their own composed mixins:
  • MarqueeSelectionMixin (src/mixins/marquee_selection.ts) — rubber-band marquee selection: updateMarqueeState_, showMarqueeOverlay_/updateMarqueeOverlay_/hideMarqueeOverlay_, getAtomsInScreenRect_, finishMarqueeSelection_. Fully self-contained, moved verbatim.
  • GroupTransformMixin (src/mixins/group_transform.ts) — group-transform pivot/centroid math (attachPivotToSelection_, computeCentroid_, moved verbatim) plus four newly-extracted helpers: handleGroupPivotChange_ and commitGroupGizmoDrag_ (the group-specific halves of the TransformControls change/mouseUp listeners — initializeEditor() now calls into these instead of inlining the logic), captureGroupDragStartPositions_ (also de-duplicates an identical inline snapshot that was repeated in beginAtomDrag_), and applyGroupDragDelta_ (the group half of the direct-drag pointer-move handler).
  • A small shared interactive_editor_constants.ts holds DRAG_THRESHOLD_PX, needed by both the base mixin's atom-drag threshold and the new marquee mixin's own activation threshold.
• Both new mixins are composed into Wave in wave.js immediately before InteractiveStructureEditorMixin (whose constructor still runs last and calls initializeEditor()/initializeSelectionRaycaster(), by which point the two new mixins' constructors have already set their own state fields to defaults).
• Deliberately not split: beginAtomDrag_/endAtomDrag_/cancelAtomDrag_ (the direct-drag lifecycle) stay in the base mixin. They're a genuinely unified state machine covering all 4 drag shapes (single/group × direct/gizmo) at once, and this round's Esc-cancel regression tests are concentrated exactly there — forcing a single/group split would trade real, hard-won correctness coverage for a readability preference. See plan/interactive-editor-spec-plan.md's "Chained PR stack" section for the full writeup.
• Base file: 1460 → 1211 lines. Fourth and last PR in the post-review chained-PR stack (stacked on #208).

Test plan

• Pure refactor — zero test file changes. All 181 pre-existing tests pass unchanged, which is the primary evidence of behavioral equivalence for a change like this.
• tsc --noEmit surfaced a real narrowing gap the split introduced (group_transform.ts's fields no longer share a class body with transformControls_/transformDragStartPosition_, so TS couldn't narrow selectionPivot_'s nullability through a compound guard referencing them) — fixed by capturing selectionPivot_ in a local before the guard, not suppressed.
• Live-browser smoke test against a real (non-jsdom, non-mocked) WebGL Wave instance: marquee-select correctly selects both atoms via a screen-space rectangle; group direct-drag moves both atoms by an identical delta; gizmo group-rotate preserves the centroid and resets the pivot's quaternion after commit.
• npm run lint clean (0 errors; only pre-existing no-explicit-any warnings matching this file's established convention).
• npm run build clean.

🤖 Generated with Claude Code

[timurbazhirov]

Closing — consolidated into #216 (then #214 on top).

Verified contained: this PR's head e918f0cf is a direct ancestor of claude/uiux-p2-figure-export.

Worth noting for TB-ARCH-2's own sake: interactive_structure_editor.ts has grown again since — U-13 adds the edit-mode touch remap to it — so the split this PR started is still worth finishing, just as follow-up work rather than as a PR waiting on the same review.

The branch refactor/split-interactive-editor-mixins stays on the remote.

---

Generated by Claude Code