Duplicate text if you run convert_to_parent_pixel_bounds or convert_to_coords twice

[rwood-97]

Describe the bug
At the moment, we store out text predictions as a list in dictionary. If you run convert_to_parent_pixel_bounds or convert_to_coords twice, it will just append the predictions onto the existing list so you will have 2x the outputs.

This should be fixed either by starting a fresh list for each time you run convert_to_xx or by converting the list to a set.
Obviously, these solutions still aren't ideal and would be better as a data class.

[rwood-97]

Maybe fixed by #536 , check!