* Do profiling and figure out what is the real slow code - parsing
  a big document with many annotations takes quite long now.
* Maybe try to preserve annotation IDs - that would help debugging
  using xmldiff. Otherwise both the original and the processed XML
  would need to be processed to remove the Id attribute before 
  comparision because those do seem to change.
  -> if using Ids we need to also assign them for freshly created
  annotations - how is this done in GATE?
* Maybe implement CompoundDocument to support those also.
* Implement offset-based accessors for annotations
* implement a fast indexing datastructure for checking for contained
  annotations: 
    do a topological sort of all annotations and put into list
    create an array for all possible offsets in the document
    for each offset point to the first ann in the sorted list which
      starts at or after this offset
    Then:
      to find contained in [a,b] find starting ann through offset[a],
      then go through list until listel.from > b and take all where
      listel.from <= b and listel.to <= b