If you use a MvNormal distribution, the default sampler should attempt to update all elements of that random variable simultaneously.
That said, my experience is that using a multivariate Metropolis proposal without NUTS or HMC is usually pretty poor in performance and scales exponentially poorly with the dimension of the RV. The default sampler isn’t going to use blockwise conjugate updates for a multivariate normal, which is how I’ve often seen handwritten MCMC implementations do it.