An infeasible-start framework for convex quadratic optimization, with application to constraint-reduced interior-point and other methods