Improve codecov reports

[chmreid]

We have experienced intermittent problems with codecov reports, mainly related to misleading code coverage statistics. For example, very significant drops in coverage (>50%) caused by changing a single line in a Readme.

The problem we have identified is that we are running five separate tests for each commit on a pull request (because there are 5 test suites that each upload codecov reports). @natanlao has proposed bumping the "report after N builds" setting to 5, so that codecov will wait until all 5 tests have finished uploading to actually compute its code coverage statistics.

This issue is linked to PR #2541 from @natanlao.

[natanlao]

No dice, unfortunately.

[chmreid]

ugh codecov is teh wuuuurst