Annex processing assumes that it runs inside Eclipse

[lwrage]

Annex parsing is implemented in a rather weird way that requires it to run in Eclipse. As a consequence, we cannot parse AADL files standalone, for example, in tests. This needs to be redesigned from scratch. Maybe #619 is the way to go.

child of #1018
blocks #1254

[lwrage]

Potential approaches to investigate:

• Check if the workbench is running and use different error reporting depending on the result
• Don't use markers but inject annex errors and warnings into xtext's way of handling them
• Use Xtext derived state computer (see Use Xtext's derived state machinery to handle annex parsing #619)
• Use an annex builder

[lwrage]

For now, try minimal changes to existing implementation:

• Use queuing error reporter for annex parsing/linking.
• Make error reporters (except marker reporter) independent of Eclipse IResource.
• Deprecate marker error reporters.
• Don't process annexes if OSGI is not available. We rely on the extension registry, even for Xtext annexes. This can probably be improved later.