[common.convergence-study] changed default 'relative' to false

This is the more natural situation if some of the norms are actually
estimators. In addition, the EOC should not be affected anyway.