Simultaneous CIs

[nhejazi]

We should implement several approaches for generating simultaneous confidence intervals covering the ATE across the full set of specified probes. A few ideas:

• Bonferroni CIs: \hat{ATE}g \pm Z{1 - 0.05/(2p)} SE_g, where SE_g is the standard error based on the pooled variance estimate

• Multivariate Normal CIs: an approach based on the variance-covariance matrix of the EIF as implemented in https://github.com/tlverse/tmle3/blob/master/R/utils.R#L55-L57