_win_print_internal now calls _win_ensure_pad_capacity() with getcury() plus the estimated message height, so deriving 'dead' from lines_needed counted that lookahead reservation as dead space. Compute it from getcury(win) directly so the reclaim keys off the real cursor-vs-buffer gap.
Variant A (PAD_THRESHOLD=12000) bounds the pad but still re-enters win_redraw() from the print path on every message once the *live* buffer renders >= the threshold — reachable with a single oversized message (huge paste, narrow terminal, crafted incoming), reintroducing the per-message redraw storm.
Trigger the reclaim redraw on dead space (getcury - buffer->lines) instead of absolute pad height. After a redraw the cursor equals buffer->lines, so dead space is 0 regardless of buffer size: the reclaim can never fire while scrolling, and an oversized message no longer storms (and is shown in full rather than clipped). Dead space only accrues in a long append-only session that trimmed old entries — exactly when reclaiming is wanted.
