'Languages are top to bottom', yes, but only
because of the legacy mechanics of handwriting and then typewriter-rolls. (Hands or mechanism would obscure any lower text.)
As each generation spends more time with keyboarded/gestural/voice inputs for flexible screens, the old mechnical habit – append new at-bottom – will fade, with more applications embracing new-item (and new-line-insert) at-top.
Regarding shells, do you mean your eye tracking would have to zig-zag? If so, imagine a completely inverted shell/terminal, at the unit of individual lines. You'd only have to scrollbar-back in the exact same situations as in a classic terminal: when more than one screen of output resulted. And your eye tracking for smaller bursts of output will require the same reversals and vertical-magnitudes as before: just in the opposite direction. So once you habituate to the inversion, the differences that remain are:
(1) 'up' means 'newer', consistent with other screen-native scrollable information feeds
(2) your eyes' 'home row' stays near the top of your screen – closer to looking at the world above the screen, rather than the hands/keyboard/ground below
Both these seem like wins to me... far from slam-dunks, but enough to chip, chip, chip away at the traditional language practices over time.