It might be useful to benchmark verification speeds of bisection & sequential algorithms.

Also, we should think if there should be a hybrid mode in the current implementation, which alternates between bisection & sequential verification depending on how far the light client is behind the latest header.