Randomly pivoted Cholesky: Practical approximation of a kernel matrix with few entry evaluations