Speed up sensitivity analysis for expanded confidence intervals and p-values.

[rwilson4]

Currently we use the approach in "Sensitivity Analysis for Inverse Probability Weighting Estimators via the Percentile Bootstrap." This is slower than I'd like. Rosenbaum has a relevant discussion in his new textbook, and Nathan Kallus has a relevant paper, "Doubly-Valid/Doubly-Sharp Sensitivity Analysis for Causal Inference with Unmeasured Confounding".

[rwilson4]

Neither of these approaches seem to pan out. Next step is to compare my heuristic from my blog post with the bootstrap-based expanded confidence intervals to see how anti-conservative it is.