From Transparent to Opaque: Rethinking Neural Implicit Surfaces with α-NeuS