Fix bug where MNIST train and test are the same

[christabella]

But maybe this doesn't matter, since both the meta train and meta test data are from generate_meta_test_data? Is the reason for this that for the synthetically generated datasets it doesn't matter?

meta_learning_pacoh/experiments/meta_overfitting_v2/meta_GPR_overfitting_base.py

Lines 75 to 79 in 7098ca2

	meta_train_data = dataset.generate_meta_test_data(n_tasks=1024, n_samples_context=FLAGS.n_context_samples,
	n_samples_test=FLAGS.n_test_samples)
	meta_test_data = dataset.generate_meta_test_data(n_tasks=FLAGS.n_test_tasks,
	n_samples_context=FLAGS.n_context_samples,
	n_samples_test=FLAGS.n_test_samples)