Currently selection in the viewport with the text tool is based on a geometry intersection. Since the geometry of the text is usually very intricate, this can be annoying.

The code for selecting existing text is in the fn interact in text_tool.rs. Currently it uses document.click(...). The new functionality would need to replace this with an iteration over the layers (from top to bottom), attempting to get their text node. The bounding box function from to_path.rs would then be applied, which would be transformed by the transform_to_viewport and intersected with the cursor.