perf(grid): eliminate per-cell allocations of BoxDecoration, Border, and TextStyle in _GridCell via flyweight cache #1519

Summary

In lib/features/workspace/grid/grid_cell_renderer.dart (lines 112–168):

Inside _GridCell.build(), which runs for every visible cell in every visible row on every scroll event:

final style = material.TextStyle(...); // Allocates a new TextStyle
final bg = isSelected ? ... : ...;
decoration: material.BoxDecoration(
  color: bg,
  border: material.Border( // Allocates a new Border
    right: material.BorderSide(...), // Allocates BorderSide instances
    left: isSelectionLeft ? ... : ...,
    top: isSelectionTop ? ... : ...,
    bottom: ...,
  ),
),

For a typical grid view with 40 visible rows $\times$ 15 columns = 600 visible cells:

• 600 TextStyle objects
• 600 BoxDecoration objects
• 600 Border objects
• 1,200 BorderSide objects
• ~3,000 layout decoration allocations PER FRAME during fast scrolling.

Because every cell gets a newly instantiated BoxDecoration reference, Flutter's RenderDecoratedBox cannot short-circuit painting via identical(oldDecoration, newDecoration).

Architectural Solution: Flyweight Decoration Pool

Cell appearance states form a finite set of combinations:

• striped (2 states)
• isSelected (2 states)
• rowStatus (clean / inserted / deleted) (3 states)
• cellStatus (clean / modified) (2 states)

1. Precompute and pool the decoration instances in a compact lookup cache (Flyweight pattern) keyed by integer bitmask:

   static BoxDecoration getDecoration({
     required bool striped,
     required bool isSelected,
     required StagedRowStatus rowStatus,
     required StagedCellStatus cellStatus,
     required ColorScheme colors,
   }) { ... }

2. Cache the 4 standard text styles (normal, nullItalic, deletedStrikethrough, modified).
3. Reused decoration instances allow RenderDecoratedBox to skip repainting when cell state is unchanged.

Acceptance Criteria

• Scrolling the data grid produces zero BoxDecoration, Border, or TextStyle allocations for clean/unchanged cells.
• 120 FPS scrolling maintained without GC stutter on large datasets (>10,000 rows).