TF2.x: How to parallelize code inside a tf.data.Dataset generator?

Question

aim is to train a tf.keras.Model with image data, where each sample is actually a stack of three RGB JPEGs, to essentially 9-channel images. Currently, my custom generator l