We need to decide on whether buffers and non-principal
documents use still the same input stack or have stacks
of their own. I found two solutions acceptable:
1. New docs have always new stacks, buffers use the
currents input stack with all consequences for matching
and accessing to a stack data. There is no way to
access other than the current stack.
2. The above rules are defaults for new docs and
buffers but the behavior can be changed with
'new-input-stack' or similar attribute to
process-buffer/document (as suggested by HJ)
3. New stacks for both docs and buffers.
Many other solutions have been discussed including
those introducing new symbols to STXPath to distinguish
nodes from buffers.