Document with a dangling relationship fails to open with KeyError

[toxicphreAK]

A document that contains a relationship whose target part is not present in the package cannot be opened at all — the failure happens during package load, before any application code can inspect or skip the bad relationship.

Reproduction

Take the bundled default template, add one relationship pointing at a file that does not exist, and open it:

# word/_rels/document.xml.rels gains:
#   <Relationship Id="rIdBogus"
#     Type=".../image" Target="media/missing.png"/>
Document("broken.docx")
# KeyError: "There is no item named 'word/media/missing.png' in the archive"

A raw KeyError from zipfile escapes the OPC layer, so a caller cannot even distinguish this from a library bug without matching on the message.

Why it matters

This is one of the most reported failure modes upstream and it is not a synthetic case. Documents lose parts through incremental save, third-party generators, DRM/repair tools, and mail gateways that strip embedded media. Word itself opens these files without complaint — it treats a relationship with no target as absent.

Scope

• Detect a relationship whose target part is missing while building the part graph in docx/opc/pkgreader.py, and drop it rather than raising
• Emit a warning naming the rId and target so the loss is visible but non-fatal
• Ensure the surviving Relationships collection stays consistent — rIds must not be dangling for later related_part() lookups
• Decide and document the behaviour for a dropped image relationship reached via InlineShape — the shape element remains in the XML with an rId that no longer resolves
• Round-trip test: open a package with a dangling rel, save it, confirm the result is loadable and the bad rel is gone
• Unit tests covering internal and external relationship targets

Notes

Upstream has three open issues and two unmerged PRs on this, the oldest from 2019:

• fix: accommodate NULL relationship (by skipping) python-openxml/python-docx#678 — fix: accommodate NULL relationship (by skipping)
• Unable to read docx containing pictures linking to internal bookmarks: KeyError: "There is no item named 'word/#MyBookmark' in the archive" python-openxml/python-docx#902 — pictures linked to internal bookmarks, same KeyError
• Open Word docx file with "The image part with relationship rID8 was not found" error, it always fails python-openxml/python-docx#1105 — The image part with relationship rID8 was not found
• PR Detect and ignore rels to missing files like Word does python-openxml/python-docx#1219 — "Detect and ignore rels to missing files like Word does"
• PR Skip null and internal bookmarks links when loading document python-openxml/python-docx#1498 — skips null and internal bookmark links

Related but distinct: python-openxml#1566 skips non-image parts when matching by sha1 in ImageParts, which is a second way a broken image relationship surfaces. Worth handling in the same pass.