Scrolling data browser and fluid workspace

Build 2026-10-24.21 replaces the data browser’s 100-row / 14-variable pages with a native two-axis scroll region. Column headings remain at the top; row numbers remain at the left. Enter/Up/Down move between rows; Tab and Shift+Tab move across variables and wrap to the adjacent row. Existing Data actions and sessions that specify a browser position still work. The grid fits the remaining window height so both scrollbars stay accessible.

Only visible rows/columns plus a small overscan are rendered. Unfiltered scrolling indexes raw rows directly; filtered scrolling builds and caches a compact index of matching original row IDs. Very large row counts use a compressed scrollbar capped at eight million pixels to avoid browser element-height limits. All rows remain reachable. This affects display only; analyses still use the full eligible dataset.

Cells expose exact raw numeric values for editing and value labels in tooltips. Leaving a cell or scrolling it out of view commits its draft through the existing edit/undo path. Escape cancels the draft. Merely focusing and leaving a cell does not write data or append rows.

The shell’s desktop maximum-width cap is removed. The workspace has 12px desktop gutters and 8px mobile gutters. Browser cells are compact (96px wide, 28px high). Browse starts with the right panel collapsed to its rail, making the table wider; users can reopen Plots, Console, Report or Help. This separate Browse preference preserves the right-panel state for other analyses. Existing responsive columns, palette, analyses and chart sizing controls remain in place.

Build: PYTHONDONTWRITEBYTECODE=1 python3 calculator/tools/build_time_series.py

Focused checks: node calculator/tests/browse-scroll.test.mjs