Demystifying the Hypercomplex: Inductive Biases in Hypercomplex Deep Learning