Estimating Neural Representation Alignment from Sparsely Sampled Inputs and Features