Loader could have streaming option

[neubig]

Currently Loader only loads the entire dataset into memory all at once. This might not be ideal for very large datasets.

It wouldn't be too hard to add a streaming option by modifying these lines to zip and yield:

ExplainaBoard/explainaboard/loaders/loader.py

Lines 148 to 151 in 08b0665

	data_list: list[dict] = output_loaded_data.samples
	for i, output in enumerate(data_list):
	dataset_loaded_data[i].update(output)
	return dataset_loaded_data

But this would also require checking throughout the codebase that the downstream usage of this function is not assuming a list, but an iterable.