In Guess Image from Pronunciation, make distractor images more diverse

[Daniel-Mietchen]

Currently, the correct answers and the distractors seem drawn from the same pool of images associated with an audio pronunciation file. For the distractor images, there is no need to limit them the same way, so perhaps they could be drawn from something more generic, e.g. any item with a corresponding lexeme, or any item.