While we like codecov, and the coverage goals, we've frequently ignored
the failures for various reasons. These have confused community members
submitting PRs and it can be confusing to look at a list of merged PRs
with the big X indicating failed checks ("confusing" isn't really a
problem, but if there were actually a bad merge, it could be hard to
find which in a list of commits actually had an important failure).
Fixes a mild inconvenience.
We have some non-deterministic tests which flap the coverage up and down
slightly, but that should not set a failing status on PRs. This commit
sets a threshold of 0.5% coverage change to prevent this.
Kristin Laemmert