GH-1405 fixed scope handling in join iterator

[abrokenjester]

GitHub issue resolved: #1405

Briefly describe the changes proposed in this PR:

• use merge-join for group patterns to handle scope, with special case
  exception for FILTER (NOT) EXISTS

[abrokenjester]

The fix seems to cause a regression in the federation sail, got two failing tests there.

[abrokenjester]

@aschwarte10 is this something you might be able to shed some light on? I've got two failing tests in the (old) federation sail due a change I made in the JoinIterator algorithm, put I can't put my finger on why. Is there an implicit assumption somewhere in the federated evaluation strategy that left argument results automatically get injected into right argument evaluation?

[abrokenjester]

This still has a number of failing compliance tests (these aren't run on PR verify) and will need more time. I think I may need to ensure we don't use merge-joining at all when the arg is a filter condition (not just when it's a filter not exists) which makes sense if you think about it: the filter needs injected variable bindings to evaluate its condition.

[abrokenjester]

Verified locally that this now passes all compliance tests. Be advised that due to the way the tests are set up, you will need to run a full mvn clean install on the compliance modules, because otherwise the federated sparql tests use an old snapshot of the rdf4j-server and consequently fail.

Another point in favor of getting rid of using embedded rdf4j servers for (most of) our test suites, and just moving to a full mock/stub approach for this kind of integration testing.