An important missing feature is a confidence interval on the delta, since insignificant run-to-run changes might wrongly show up as meaningful.

This PR adds that. Notice the cache_misses - 3.3% (which would've been considered significant before) is no longer significant since the standard deviation was high.

Related to #7 and #2

Resources:

• T-table came from the two-tailed 95% column here: https://en.wikipedia.org/wiki/Student%27s_t-distribution.
• Underlying stats described here: https://sphweb.bumc.bu.edu/otlt/mph-modules/bs/bs704_confidence_intervals/bs704_confidence_intervals5.html