Evidently, this wasn't totally fixed. Apparently, setting the cursor twice on multi-monitor setups is necessary, see: https://stackoverflow.com/questions/65519784/why-does-setcursorpos-reset-the...
The other issue is that sometimes rarely on Windows, it seems it doesn't autocomplete, but we use a sleep to check for it (it's not possible to check reliably, unfortunately). So as a workaround, I've increased the sleep but only if it fails the first time.
Hopefully these new commits Are a lot more robust. They don't fix all the rare failures, though.