School Choice as a Latent Variable: Estimating the “Complier Average Causal Effect” of Vouchers in Charlotte