docs(internet): convert bare citations to the issue role in docstrings

[JarryShaw]

Please follow the guide below

• Searched for similar pull requests
• Followed the coding style
• make test passes, and a test case covers the change — docstring-only, so no test change, matching docs(corekit): convert bare citations to the issue role in corekit docstrings #1000
• Added a changelog entry — N/A, centralised in docs(changelog): shared 1.5.0 changelog — long-lived, merges last (#610, #616, #617, #618, #620) #657

What is the purpose of your pull request?

• docs — documentation only

---

Description of your pull request and other information

Fourth tranche of #989, alongside #1002. Converts bare #NNN citations under pcapkit/protocols/internet/ and pcapkit/protocols/protocol.py to :issue: roles, continuing #1000's corekit pass.

Scope follows the ruling on #989: plain # comments keep the bare form, so only docstrings and #: autodoc doc-comments convert.

Tokenised by token kind, base e8a60d153 → head:

	base	head
bare in strings	115	36
bare in #: doc-comments	5	0
bare in plain # comments	58	58, untouched
:issue: roles	0	84

The 36 survivors are RFC packet-diagram labels, not citations — DH GROUP ID #1, HIT #1, Suite ID #2 and the like, all in hip.py. Every one is a single digit, and no 3-or-more-digit bare citation survives, so nothing real was skipped on that boundary.

84 roles across 25 distinct numbers, every one an issue rather than a pull request. 79 of the 84 sit inside docstrings (ast-checked); the other 5 are the #: doc-comments, which ast cannot see. None is a runtime string, so no message text changed, and the #%s caption leaves the rendered text identical. Docs build stays at 61 warnings and the two pre-existing docutils errors.

[JarryShaw]

GOOD TO GO at 6f59ff9c5 — opus cross-review, a different model from the sonnet agent that wrote the diff, briefed to falsify. No code defect. It found one real error in my own description, which I have corrected above, and it overturned a correction I had made:

• I wrote that all 84 role sites are docstrings, ast-checked. Only 79 are. The other 5 are #: attribute doc-comments, which ast cannot see at all, so the check I claimed could not have covered them — and the body's own table lists those 5 separately, so it contradicted itself. Re-derived: 79 roles in string tokens, every one in docstring position, 5 in comment tokens. The substance holds — no role sits in a runtime string — but the wording did not.
• My note that the authoring agent's "39 distinct numbers" was wrong is itself wrong. Both counts are real and measure different sets: distinct 3-or-more-digit citations across all token kinds, including the 58 plain comments left alone, is 39; across strings and #: only — the converted set — it is 25. Measured both on e8a60d153.

Independently confirmed beyond that: all 36 surviving one-digit citations sit inside .. code-block:: text RFC diagrams in hip.py, 0 outside; a prose-cue scan for short citations in prose returned 0 hits, so nothing real was skipped on the 3-digit boundary; the 58 plain-comment citations match base on file, number and comment text; 25 numbers, 25 issues, 0 PRs; and a base-vs-head comparison with every role collapsed back to #NNN is identical on all 11 files, which proves the diff is pure markup.

One thing worth knowing for the rest of #989: two GH-NNN citations at protocol.py:926,970 survive, both in plain # comments and so out of scope under the ruling. GH- is a spelling the sweep's #NNN framing does not cover.

Not ready to merge yet — CI still finishing, 0 failures so far.

[JarryShaw]

Ready to merge at 6f59ff9c5 — closing the "not yet" on the verdict above. CI is complete: 69 CheckRun legs green, 3 skipped, 0 failures, 0 in flight, mergeStateStatus CLEAN. The opus verdict stands at this head, and the description fix it prompted is in place; nothing has been pushed since.

Yours to merge.