assign_and_map_hash_ids() reuses the same id for two elements of a page that is not contiguous #4620

Describe the bug

unstructured/documents/elements.py derives the "sequence number on page" with itertools.groupby(page_numbers), so the number is the index inside each contiguous run of equal page numbers and it restarts when a page number appears again later in the list. The comment above the code ("generate sequence number for each element on a page") and Element.id_to_hash() ("sequence number on page") both describe a per-page counter, and the sibling _assign_hash_ids() in unstructured/partition/common/metadata.py implements exactly that.

The visible effect: two distinct elements that share their text on a page that is not contiguous in the list get the same id, so old_to_new_mapping silently points both parent_ids at one element.

To reproduce

import copy
from unstructured.documents.elements import ElementMetadata, Text, assign_and_map_hash_ids

elements = [
    Text(text="Section", metadata=ElementMetadata(page_number=1)),
    Text(text="Interlude", metadata=ElementMetadata(page_number=2)),
    Text(text="Section", metadata=ElementMetadata(page_number=1)),
]
ids = [e.id for e in assign_and_map_hash_ids(copy.deepcopy(elements))]
print(ids, len(set(ids)))  # 2 unique ids for 3 elements

Expected behavior

Three unique ids: the third element is the second element of page 1, not the first one again.

Version info

Observed on main at 53a56a0a (unstructured 0.27.35); no extra packages are needed for the snippet above.