Update hydrological states or meteorological forcings? Comparing data assimilation methods for differentiable hydrologic models