[compare.py] Add confidence interval (#377)

This patch adds a `--diff-confidence-interval=relative|absolute` option to `compare.py` to report 95% (or 1-alpha) confidence intervals for the relative difference between `lhs` and `rhs` runs.

The current p-value and significance markers only tells the user if a difference is statistically significant against null hypothesis, but does not show how large the true difference might vary.

Example output from `compare.py ... --statistics --diff-confidence-interval`
```
     Program                                       exec_time
                                                   lhs       rhs    diff  std_lhs std_rhs t-value p-value significant diff_ci_rel
     C                                               2.95      3.40 15.3% 0.076   0.100   -6.653  0.0027  Y           [ 9.3%, 22.7%]
     A                                               1.00      1.15 15.0% 0.050   0.050   -3.674  0.0213  Y           [ 3.5%, 25.1%]
     B                                               1.95      2.20 12.8% 0.076   0.050   -4.427  0.0114  Y           [ 4.3%, 18.8%]
                                Geomean difference                  14.4%
```