Revision 21007

Fix for bug 1264 from Julian Cable:

    I've refactored the offending code branch and added some comments so
  hopefully the intent is a bit clearer. The loop termination conditions
  are now obviously independent of the content on the wire (they were
  meant to be before, but I admit it was obscure). I've tried using the
  ephemeral memory routines.

Add a check for a maximum fragment count, and bail out of reassembly instead
of triggering an ep_alloc exception.  Add Julian to AUTHORS.  Update the 
release notes.