(Non)uniqueness of minimizers in the least gradient problem