Circular dependency detection: Outputs which are pogrammatically set to faulty are treated like external inputs.

- Implemented but not tested yet
- All existing tests passing

* Not implemented yet. Currently always return false.
* Test exists and compiles, but is failing due to the missing implementation.

Paths starting with "/" will be stripped of the leading slash now. Empty paths (or a single slash) will be interpreted to refer to the Module itself and hence return *this.

This logic is otherwise ofter required in code which uses submodule().

This actually makes the implementation also easier and more efficient.

* add HierarchyModifyingGroup

* VirtualModule: allow nested oneLevelUp etc.; simplify implementation

This is achieved by properly setting the _owner of each VirtualModule (unless highest level).

The simplification comes from then being able to use getOwner() instead of keeping track of "grandparents", and from using createAndGetSubmodule() instead of implementing something similar again in findTagAndAppendToModule().

* DataLossCounter constructor now takes qualified path to trigger variable

Co-authored-by: Jens Georg <jens.georg@desy.de>

Prints a clear message if a circular dependency of initial values exists.

fix tsan complaining about incrementing/decrementing testableMode_deviceInitialisationCounter outside testable mode

- Also removed race conditions in the test.

- Version number in exception case is always the same.
- Remembers that it has reported an exception so a readLatest() after recovery does not switch the data validity back to ok if there was no new data.

- All variables conceptually have an initial value now.
- All initial values read by blocking read() before starting the main loop.
- More tests failing, probably to old test exceptiations and wrong test implementations.

- tests still failing
- modifications incomplete
- cleanup needed

- Initial values are sometimetimes only propagated in the first run of the main loops, not in prepare. If other modules are waiting for it they cannot enter their main loop. That's why all threads have to be started first, and only the it is waited for all of them to enter testable mode.