Clients will need to request an update operation
on model elements; we should provide a mechanism
for that. The updater implementation may decide,
based on timestamps or some cached `dirty' status
to perform or discard the request, coalesce
requests or whatever. The update operation should
be  asynchronous; client will get Task handle so
it can wait for the operation to  complete.
Initially, it will be sufficient to update whole
sources or classes; after we get an incremental
parser, we may be able to do smaller update
operations.