Text this: RetiNerveNet: using recursive deep learning to estimate pointwise 24-2 visual field data based on retinal structure