Fix stale rows when scrolling a restricted region (CoreGraphics renderer)

[tavvet]

Problem

Full-screen ncurses apps that keep a fixed header and/or footer — nano, and vim/less/htop in some layouts — set a restricted scroll region (DECSTBM, e.g. rows [1, 43] of a 46-row screen). While scrolling such a region, rows outside the region keep stale pixels, and a "ghost" line lingers on the very last row while paging.

Reproducible on the CoreGraphics renderer (Metal disabled): open nano on a file with lines wider than the window and hold PageDown/PageUp — stale rows and a bottom ghost appear.

Root cause

scroll(), reverseIndex(), cmdScrollUp() and cmdScrollDown() flag only [scrollTop, scrollBottom] for refresh via updateRange. For a restricted region (scrollTop != 0 or scrollBottom != rows-1) or the alternate buffer:

• the rows outside the region (and the whole-viewport shift scroll() makes via yBase) are never repainted, and
• the last row is only repainted when getUpdateRange() returns rowEnd == rows-1, which triggers updateDisplay's bottom-edge "remaining bits" branch — a restricted region never produces that rowEnd.

So those rows keep stale backing-store pixels on the CG path.

Fix

Add refreshScrolledRegion(top:bottom:): when the region is restricted or we're on the alternate buffer, repaint the whole viewport (endLine == rows-1); a real full-screen scrollback scroll keeps the cheap region-only path (its scroll blit handles the bottom edge). Call it from the four scroll primitives. The whole-viewport repaint is safe on the Metal path too (it just widens the dirty range).

Testing

Logged updateDisplay's (rowStart, rowEnd): a restricted-region scroll now produces rowEnd == rows-1 (bottom-edge branch) instead of rowEnd == scrollBottom. Verified visually in nano/vim/less — no stale rows or bottom ghost while paging. The package builds cleanly.