Lower limit on R squared with two weak regressors

OLS regressions of $y$ on $x_1$ alone and on $x_2$ alone (each with an intercept) both have $R^2=0.05$. What is the largest value that is guaranteed to bound the $R^2$ of the regression of $y$ on both $x_1$ and $x_2$ from below?

$0.05$. Including an extra regressor cannot reduce the explained variance, so the joint $R^2$ is at least $0.05$; it equals $0.05$ when, for instance, $x_2$ is identical to $x_1$ and adds no information.