We disallow concept nodes in annoations / annotation instances in containments

[enikao]

In #317, we started a discussion whether we want to allow nodes with concept classifier in annotations, or nodes with annotation classifier in containments.

NOTE 1: This is a validation issue, not a serialization/protocol issue (in first approximation).
Both bulk and delta protocol are language-agnostic (#226 / #300). Without knowing the language, we cannot tell whether a classifier is a Concept or Annotation.
It influences delta protocol insofar as we might have (or have not) special commands to move nodes between annotations and containments.

NOTE 2: This only applies to nodes directly inside the Host. Both a sub-node with Annotation classifier and a sub-node with Concept classifier can have annotations and containments on their own. Example: A Documentation annotation can have lots of children of Concept Word.

NOTE 3: We always have to adhere to the type of a containment. So a containment of type IMember can never host any sub-node of a classifier that's not a specialization of IMember, irrespective whether that sub-node is of a Concept or Annotation classifier.
Similarly, we always have to adhere to the annotates specification of an Annotation.

Example:

M2:

Concept Host
  containment members: IMember[0..*]

Interface IMember

Concept ChildMember implements IMember

Annotation AnnotationMember implements IMember
  annotates Host

M1:
MyHost <<Host>>
  annotations:
    AnnA <<ChildMember>>
    AnnB <<AnnotationMember>>
  containments:
    members:
      ContC <<ChildMember>>
      ContD <<AnnotationMember>>

Option A: Allow concept nodes in annotations and annotation nodes in containments

Pro:

• Most flexible

Con:

• Concept nodes would "fit" to more parents than annotations, as we could only assume a concept "annotates" INode, i.e. everything.

Option B: Allow annotation nodes in containments

Pro:

• Does not violate containment or annotates types
• Can support edge cases, e.g. we attach a Documentation node as child to all concepts that are under our control, but as annotation to concepts we use from a third-party (i.e. unchangeable) language.

Con:

• Still requires mechanisms to deal with nodes outside their "natural habitat"

Option C: Disallow concept nodes in annotations and/or annotation nodes in containments

Pro:

• Keeps the original intention of both annotations and containments: Containments are integral part of a Concept, annotations are orthogonal / auxiliary aspects.
• Simplifies delta protocol
• Might introduce issues in deserialization
• Checking is not hard

Decision: Option C

Con:

• Gives up some flexibility