A small-gain-theorem-like approach to nonlinear observability via finite capacity channels*