This is a known problem and happens due to a combination of complex webpage layouts and using touch as input.

The thing with touch input is that it's not precise at all. Firefox routinely gets told coordinates which are not quite where the user intended. With normal taps, it looks around for the nearest interactive element and activates that. But with text selection, there is no interactive element to use for orientation.

That's not a problem when a webpage is structured like a document. Then the small imprecisions will just mean that one letter more or less is selected.
But if the webpage is structured in such a way that elements, which are visually next to each other, are actually far away from each other in the HTML structure, that's when it becomes a problem.
The text selection follows this HTML structure, but the touch interpretation needs to follow the visual layout, since you may very well want to extend the selection into another HTML element.

And so, because it's not quite sure which pixel you're touching, and there's pixels next to each other which are miles away in the HTML structure, the text selection just jitters around like crazy.