docs: tighten demo.rst and the docs landing page (#719)

[JarryShaw]

Please follow the guide below

• Searched for similar pull requests
• Followed the coding style (make pylint, make mypy, make isort)
• make test passes, and a test case covers the change
• Added a changelog entry under docs/source/changelog/ and regenerated CHANGELOG.md, if the change is user-visible — N/A — documentation prose only, no library behaviour change

What is the purpose of your pull request?

• fix — corrects a defect
• feat — adds a feature
• perf — changes performance, not behaviour
• refactor — changes neither behaviour nor performance
• test — tests only
• docs — documentation only
• ci — workflows or build tooling
• chore — anything else

---

Description of your pull request and other information

Second slice of #719, taking the two user-facing root pages #953 did not touch: docs/source/demo.rst and docs/source/index.rst. +29/−32.

One real documentation defect, at six sites. All three CLI transcripts in demo.rst showed 🚨Loading and 🍺Report with no space. The space is real: pcapkit/__main__.py:135,143 emojizes ":police_car_light: Loading file ..." with the space inside the f-string, and emoji.emojize substitutes the shortcode in place, so the output is 🚨 Loading file .... Confirmed by running all three invocations against examples/captures/in.pcap and reading the raw bytes.

One claim restated against the table it cites. Engine Comparison said pcapkit's ~0.2 ms/packet was "not enough comparing to other popular extraction engines", which the Test Results table does not support — pcapkit at 0.2342 ms is slower than pypcapfile, dpkt, pypcap, pcap_ct and scapy, and roughly eighty times faster than pyshark at 18.8158. It now says slower than most of the engines below.

Timed context removed: "introduced alternative extraction engines" and "By now pcapkit supports …", both of which date prose describing the present. Three typos went with it — foundamental, the gi repository, and a missing relative pronoun in the jspcapy note.

Untouched, deliberately: every toctree entry, the .. deprecated:: 0.8.0 directive, the bpc-poseur / pypcap-versus-pcap-ct / all-extra notes, and the preflight-check rationale. Those are design decisions a reader would otherwise undo, which #719 explicitly protects. docs/source/changelog.rst is excluded — its treatment is still an open question.

Verification. All three demo.rst Python examples and all three CLI examples re-run against the real capture; extract()'s keywords checked against pcapkit/interface/core.py:73-86; the engine literals against pcapkit/foundation/extraction.py:74-75; the eight module-structure parts, the pcapkit-cli script name and the pcap-ct/libpcap pins against pyproject.toml. tests/project/test_conventions_doc_claims.py plus its two siblings: 44 tests, 1 skipped, 0 failed, re-derived under plain unittest as well as pytest.

Not verified: the Test Results benchmark figures themselves and the bpc-poseur upstream line reference, both out of reach without installing engines or that package. Neither was edited.

[JarryShaw]

GOOD TO GO at 8a7b5625f — fable cross-review, read-only and on a different model from the author.

It verified the emoji fix the hard way rather than by reading the source: ran all three CLI invocations and hexdumped the output, getting f0 9f 9a a8 20 4c 6f 61 64 69 6e 67 — 🚨, 0x20, Loading. Then it went further than the fix and diffed all 24 transcript lines against the real stdout of the three runs, byte for byte, empty diff. That is the check I had not made: correcting one line of a transcript while a stale line sits beside it would have been worse than leaving both.

On the speed claim it reached the same conclusion independently — five third-party engines faster than pcapkit's 0.2342 ms, pyshark ~80× slower — and made the sharper point that the old wording implied pcapkit trailed all of them, which the pyshark row contradicts. So the new sentence is a correction, not just a shorter one.

For the removals it grepped every cut line for toctree, :ref:, :doc:, .. version*, .. deprecated::, anchors and substitution definitions: zero matches. The one :mod:pcap`` reference that went is covered by the untouched paragraph above it, which is what the reworded back-reference now points at.

Two nits accepted but not acted on, and I would rather say so than quietly fold them in:

• "can be reached two ways" wants "in two ways". Real, but a new head invalidates this verdict and restarts 61 CI legs for one preposition.
• the jspcapy note's trailing clause still attaches ambiguously to pcapkit. Pre-existing, and not worsened here.

One scope difference worth naming rather than calling an error: the body says 44 tests over three files, the review got 43 over two. Both are right at their own scope — it did not count test_documentation_claims.py, which reads neither file.

Unverified by either of us: the Test Results benchmark figures and the engine-availability matrix. Neither is edited by this change.

Labelled review: good-to-go. Not calling it ready to merge until CI finishes — and it is unpublished-equivalent in the sense that matters here: yours to merge, not mine.