[PonteIneptique]

A citeStructure such as match="/TEI/text/body/div" with a child matching ".//milestone[@Unit='section']" was unusable: the same @n value occurs in every chapter, and the parent path is absolute, so nothing scoped the child to its own parent.

• normalize_xpath: drop the "." step a ".//" child match produces, which made reconstruct_doc copy the parent twice (
  ).
• reverse_ancestor: that same "." step generated the invalid "[ancestor::]" (XPST0003).
• _get_sibling_xpath: bind $end with head(), several siblings could match the boundary xpath and "<<" rejects a sequence (XPTY0004).
• get_passage: always prefer a concrete boundary node over a tag/attribute xpath fragment, which otherwise matched a later chapter's section and over-copied. _prune_at_node prunes at a node, and _treat_siblings excludes the sibling that contains the boundary from the "<<" copy (it starts before it) to copy it pruned instead.
• get_next_unit_boundary: a unit with no next sibling fell through to COPY_UNTIL_END, returning the rest of the document (157KB for one section). It now stops at the next unit of the same kind, clamped to the end of its own parent so a section never reaches into the next chapter.
• find_refs / _resolve_groups: a child match written "//x" joined to ".///x"; treat it as ".//x", like generate_xpath already does.
• find_refs: raise a located error on an empty citation value instead of "Reference '' does not match the expected format." further down.