Asymptotic behavior of $\ell_p$-based Laplacian regularization in semi-supervised learning