How high is `high'? Rethinking the roles of dimensionality in topological data analysis and manifold learning