Make `diffcov` test smarter
Right now, diffcov runs compares the coverage of the pull head against the coverage of the master.
This doesn't make sense in many scenarios, like:
- MR targetted to release-3.1 should not compare coverage with master
- Needs master branches of the forks to always be updated
- Takes a lot of time to run redundant tests (2 times for each python version, on top of other tests)
What would be ideal?
- Store the master's coverage.ini as an artifact in gitlab and compare the branch's coverage to that. This might be a problem for forks, and also needs the master branch's coverage report to be present.
- Remove sqlite tests as coverage tool will run tests with sqlite anyway