Support batching sequence labels pairs for zero shot classification pipeline

### Feature request Currently zero shot classification pipeline dose not support specifying different set of labels for each sample input when batching multiple sequences of input. We want support input of a pair of sequence and its labels so that we can run prediction on them in batch. Following is an example of what sample input we want to specify: ``` samples = [ ("Iphone 15 pro max", ["mobiles", "appliances", "toys"]), ("Iphone 15 pro max", ["Samsung", "Apple"]), ("galaxy s21 ultra", ["mobiles", "appliances", "toys"]), ("galaxy s21 ultra", ["Samsung", "Apple"]) ] ``` ### Motivation By grouping labels with sequences pairs we can batch different type of labels for each sample and by batching it runs on gpu will be faster. ### Your contribution I will work on this feature and create a pr for it then update this issue. Making the issue now just to make sure that no one is working on this while I am doing it.