Confidence Intervals, Duality and the Bootstrap

Two variants report average revenue per user with 95% CIs: A: 4.10 (3.70, 4.50), B: 4.60 (4.15, 5.05). n = 2,000 per group. Revenue is zero for 90% of users and heavily right-skewed for the rest.

1. A colleague says "the intervals overlap, so the difference is not significant." Evaluate this claim; compute the approximate CI for the difference and the corresponding p-value.
2. State the duality between confidence intervals and hypothesis tests, and how you would use it to read the result in part 1.
3. Given the skew and mass at zero, describe how you would compute a bootstrap CI for the difference in means and for the difference in medians, and comment on whether the median is a sensible metric here.