Epic: Buffered Grid - High-Performance Locked Columns (locked: 'start' | 'end')

Input from Gemini 3.1 Pro:

✦ Epic Status Update: Architectural Foundation Complete

All core architectural sub-tasks for the high-performance Locked Columns feature have been successfully implemented and pushed.

What we accomplished:

1. Mathematical Layout Engine (#9457): The Grid now inherently sorts the columns array (start -> unlocked -> end) before VDOM generation. The passSizeToBody engine was refactored to calculate logical X coordinates purely mathematically for fixed-width grids, gracefully falling back to DOM measurements only when dynamic sizing (flex) is used.
2. Synchronous Pinning Addon (#9458): We created the GridColumnScrollPinning Main Thread addon. It maps horizontal scroll events synchronously to --grid-locked-start-offset and --grid-locked-end-offset CSS variables. The registration is completely dynamic, evaluating hasLockedColumns to ensure zero overhead for standard grids.
3. Cell Pooling Bypass & SCSS (#9459): grid/Row.mjs was split into strict passes. Pass 1 skips locked columns. Pass 2 forces locked columns to render as "Permanent Cells", preventing them from being unmounted/recycled when scrolled out of view. The transform: translateX(...) offloads the visual shift to the compositor, avoiding layer explosions.
4. Drag & Drop Integration (#9460): The SortZone interface now infers locked states based on drop boundaries. We built a robust pipeline (grid.Container#onColumnLockChange) that fully handles sorting, DOM syncing, and heavy layout recalculations seamlessly across the framework without duplicating logic.

Next Steps (Upcoming Phase):

• Unit Testing: Add tests to verify the pipeline logic and array sorting.
• Example Integrations: Integrate locked columns into existing robust examples.
• New Feature Demo: Create a dedicated Locked Columns example app.
• E2E Automation: Validate the visual compositor transforms and drag-and-drop boundary inference via Playwright.