Constant-gain nonlinear adaptive observers revisited: an application to chemostat systems