[libc++] Major improvements to visualize-historical

This patch moves to using pandas.DataFrame for representing the data,
which is a lot more expressive and better suited to computations than
regular Python objects.

- Instead of producing line charts, produce a scatter plot with
  trendlines, which is immensely more useful due to the inherent
  nature of historical benchmark results, which contain noise.
- Greatly reduce the duration for sorting commits by using `git rev-list`
  and passing all commits at once instead of requiring one subprocess
  call for each comparison.

0 files changed, 0 insertions, 0 deletions