I think this may have accidentally been changed in the refactoring.
If we discard self._rows, it will result in the entire table flashing
"..." until the new data is available. Instead, we leave the cached
rows alone, and just update the cutoff point, so we can serve stale
content (avoiding any visible redraw) until the new data is available.
I've updated search() to reset the rows there, so we free up memory
on a new search.
The backend knows exactly which op has executed, and it saves us having
to re-implement this logic on each client.
Fixes the browser table refreshing when toggling decks.
Instances can pass handled_by=self to more easily ignore events they
initiate.
Fixes ugly refresh when expanding/collapsing decks, but we're still
refreshing the card/notes area unnecessarily in that case.
1) Check whether full row height is in viewport instead of just the
top left corner.
2) Add timer before scrolling to current row so editor will already
be set up.
