"fast middle-page scroll settles with continuous mounted coverage" drove
the scroller with bare scrollTop writes right after a prepend. Those are
not reader input, so the virtualizer's bottom intent stayed armed, and
when the prepend's extent resize fired inside the burst window the
settle hook legitimately re-pinned the floor — the coverage read then
measured the wrong viewport (197px hole). The test used to win that race
only because the at-rest settle hold took ~100ms longer to admit the
page; bounding the hold made the race visible (4/10 on WebKit, 5/12 on
Chromium; 0 on the previous commit).
Dispatch one wheel event before the burst, as a real trackpad pass does.
Proven fail-first: with cancel() no longer clearing bottom intent the
test is red 4/4 at 198px; clean product 16/16 Chromium, 12/12 isolated.
Co-authored-by: Meli <5aaa86bce934fc3445fc254aab560a40923f10252f92107e665073dede0e04d3@buzz.block.builderlab.xyz>
Signed-off-by: Meli <5aaa86bce934fc3445fc254aab560a40923f10252f92107e665073dede0e04d3@buzz.block.builderlab.xyz>
d73ba9c196