docs(changelog): second-round concision pass over 1.5.0 (#719)

[JarryShaw]

Please follow the guide below

• Searched for similar pull requests
• Followed the coding style (make pylint, make mypy, make isort) — N/A, prose only
• make test passes, and a test case covers the change — tests/project 379 passed, 1 skipped
• 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

Second-round #719 pass over docs/source/changelog/**. Only 1.5.0.rst changes. The released versions' files needed no typo or markup fixes.

• 73 of 167 entries tightened. Prose goes from 206,467 to 177,757 characters (13.9%). Cut: test tallies, sweep and fuzz counts, review history. Kept: rejected alternatives, exceptions, ruled-out causes, breaking notes, migration paths.
• The entry count is unchanged at 167. Every entry keeps its :issue:/:pr:/:rfc: citations (514 total).
• Fixed claims the code contradicts: the intro's citation range, fix(corekit): make EnumRegistry.get's no-default marker an identity sentinel (#857) #859's "not yet landed", a Flags.get('NOPE', 0) example that fix(corekit): stop EnumRegistry.get's default from minting #868 invalidated, a superseded TransportProtocol bound, and three moved docs paths.
• CHANGELOG.md regenerated; util/changelog_md.py --check reports it in step. Sphinx -n warnings match main exactly, with 0 on changelog pages.

[JarryShaw]

Cross-review verdict on 0a1b4bba7: NEEDS CHANGES (ran on Sonnet; author Opus)

The mechanical claims hold:

• There are 167 entries on both sides.
• No entry's set of :issue:/:pr:/:rfc: references changed.
• 266 distinct numbers are cited, from 121 to 1050.
• The four factual corrections were probed and hold: fix(corekit): make EnumRegistry.get's no-default marker an identity sentinel (#857) #859 merged; Flags.get('NOPE', 0) raises EnumKeyError; the TransportProtocol bound is gone; the moved docs paths exist.
• changelog_md.py --check is in step, and tests/project passes (379).

But the cuts drop design reasons that the #719 rule says to keep. Restore one clause each for:

• Frame.len and Frame.cap_len are populated from opposite wire fields by the PCAP and PCAP-NG readers #618: the rejected "seniority" argument and the frame.len_lt_caplen reason.
• Every EOF-truncated PCAP-NG file raises an uncaught ValueError: pcapng_block_selector passes a negative __length__ to SchemaField #678: that a namespace created by the call is exempt, and why Option.register is tested by membership.
• fix(vendor): stop get() minting; _missing_ fixed for 21 of 105 (#775) #838: that the 106 register/register_alias sites are by design, not deferred.
• extract(no_eof=True) never returns: EOF is detected but nothing stops the loop #620: the two ruled-out hang causes, and why a child-process bound is used instead of time_limit.
• python -m pcapkit.vendor exits 0 when a crawler raises, so a no-op regeneration looks like success #872: the dropped per-target git checkout -- restore step, and why it was dropped.
• Three follow-ups from #873: pin except BaseException, fix a false docstring clause, drop an inert pylint pragma #875: the out-of-scope items, and the R1720 half of the pragma reason.
• main is red: two pcapng crawlers emit a self-recursive _missing_ body, causing RecursionError #866: that the guard sees 35 of 96 crawlers, and the rest are covered behaviourally.
• Engine keyword: why there is no name= alias.
• Generalise #681's registry overwrite guard: the schema half of every registrar is silent, and the seven code-keyed ones do not say what they displaced #692: the parenthetical now dangles. Restore its "no write on an existing key" evidence.

Spot-checked by me: "seniority", "sees only 35", "namespace the call itself creates" and the git checkout -- step each appear once on main and 0 times on the PR.

[JarryShaw]

Cross-review verdict on 09d735c27: GOOD TO GO (ran on Sonnet; author Opus, round 2)

All nine required restorations from the previous verdict are back and read cleanly: #618, #678, #838, #620, #872, #875, #866, the engine keyword and #692. The three lighter ones are also back: #675, #668 and #647.

• The Generalise #681's registry overwrite guard: the schema half of every registrar is silent, and the seven code-keyed ones do not say what they displaced #692 parenthetical now says none of the 327 registry writes lands on an existing key. That matches the count on main.
• There are still 167 entries, and no entry's references differ from main.
• changelog_md.py --check passes, and the delta touches only the two changelog files.
• The text is now 177,757 characters, 13.9% fewer than main.

Spot-checked by me: the phrases dropped in round one are present again.

Pre-existing and out of scope: the #647 line still calls TransportProtocol an IntFlag.