Improve zeta for large positive Re(s)

[fredrik-johansson]

The Riemann zeta function and Hurwitz zeta function are very poorly implemented for large Re(s). When this exceeds a fraction of the precision, one should use the Dirichlet series or the Euler product directly. Currently this is only done for machine-size integer s, but it should clearly be done generally. To begin with, zeta could call acb_dirichlet_l_euler_product in appropriate cases.