>it's a real hard-coded limit
It's 2D Data. There's a million different space optimizations you could do trivially. What you're actually limited by is rendering (but if that's an issue they already have major issues because who really has 1024x1024 cells on screen to begin with) and potentially wasting computations in your row wide operations. But any programmer should see the solutions to that. Split columns into groups and flag groups as unpopulated would be the quick dirty way.
I understand open source software has limited development resources but to me this seems more like a 'our code base sucks'-issue than a 'lack of time due to the native complexity of the task'-issue.
Slightly more perplexing than the 1024 limit is how they're bounded by 2^20 on the row axis. Why 2^20? Its like they've got no clue what they're doing. They just aimed for a billion for some reason. So arbitrary. >>63364301
You'll easily run into 1024+ columns in the most basic 'data science' (i hate this label for code monkeys) tasks. It's popular to use spreadsheet tools to get some basic grasp on the data.