Improve dependency handling in clustering/infinispan subsystem tests

The subsystem tests in clustering/infinispan are not configured to use a controller operating in --admin-only mode, but they are also not doing all the needed setup to have needed services and resources (core, jmx, jgroups) available for when the services they add are started.

This is fragile, because all sorts of stuff is failing when things are getting launched, but the tests are not noticing that. That's ok right now, but it could lead to missing regressions.

Also, the changes I'm making for WFCORE-102 mean these problems will no longer go unnoticed and the tests will fail.

I plan to:

1) Shifts tests to using --admin-only in cases where the tests are clearly not testing anything related to runtime execution; i.e. they are just testing model.

2) For OperationSequencesTestCase and OperationsTestCase, where there is some validation of runtime behavior going on (explicit in some places; in others perhaps expected, perhaps not) I'm going to switch the tests to a more focused config file that:

a) Only uses local caches, to avoid pulling in requirements for jgroups things that are not present. (I see no indication any of the tests are testing anything not present with local caches.)

b) Uses start="LAZY". This means services will get registered, but not started. Not starting avoids detection of various missing dependencies, including a ModuleLoader dependency that I can't figure out how to satisfy.

The tests can't be trying to validate any behavior when services start, because they've never been able to start. So LAZY is fine. Some tests do seem to be looking for problem related to service conflicts as resources are added/removed. These tests are still valid, because service conflicts happen as soon as services are installed, whether or not the services need to start.