Seen with PDB generated with "latest" MS compiler.
The data here is likely to be in the same form of other hash structures we already handle. (eg has a first vector to describe present entries, and a second one to describe deleteid entries; we were incorrectly expecting an empty deleted vector... so just skip for now the deleted vector in winedump and dbghelp... this will probably end up with a more generic code for reloading such hash structures, but after code freeze).