Breaking change in behavior for implict nodetype calculation

During extensive validation of backward compatiblity of droping in head (3.5.1-SNAPSHOT) over 3.4.0 led to detection of a breaking behavioral change.

The simplified case looks like this:

• a node is already present as nt:unstructured
• a package is being installed with an the filter root not having a .content.xml
• the child node of this node would be nt:unstructured

Experienced outcome in 3.4.0: all 3 nodes would be of nt:unstructured
Behavior in 3.5.1-SNAPSHOT: ConstraintViolationExceptionas no matching NodeTypeDefinition is found for the child node (indicating that the parent wasn't created as nt:unstrructureed)

The original case is slightly more complex where the substructure is deeper with some excludes and an "implicit" node (not covered by the filter due to the exclude would fail to create with the same ConstraintViolationException.

My current assumption is that solving the provided testcase (will create IT for the PR) would also solve the more complex scenario - if necessary I can add a secondary test case closer to the real-world failure.