Seol mar théacs é seo: A Riemannian perspective on matrix recovery and constrained optimization