Don't use goacc on master branch

Created by: eseliger

Currently, all PRs report a decrease in coverage, which creates a lot of noise and breaks the trust in our coverage tracking. The root cause is that we use two different methods to determine the coverage, on master we use go-acc and on branches just go test -cover. Since codecov uses the merge-base on the master branch as a base for the coverage comparison, there will always be the difference that these two algorithms create. Since go-acc is around 4 times slower than go test -cover, I think we should (for now) just use the less accurate one.

Closes https://github.com/sourcegraph/sourcegraph/issues/12103