test(generators): HTTP/2 sample frames are all written on stream 0

[JarryShaw]

Please follow the guide below

• Searched for similar pull requests
• Followed the coding style (make pylint, make mypy, make isort) — ran pylint/isort on the two files only; no new findings on my lines, make targets not run
• make test passes, and a test case covers the change — only the modules listed below, not the full suite
• Added a changelog entry under docs/source/changelog/ and regenerated CHANGELOG.md, if the change is user-visible — N/A, added centrally after the wave

What is the purpose of your pull request?

• test — tests only

---

Description of your pull request and other information

Closes #1156. _httpv2_build now puts SETTINGS/PING/GOAWAY on stream 0 and every other frame type on stream 1; PUSH_PROMISE promises stream 2. The Length convention needed no generator change: on 2c7a97c84 the frames already declare the payload length (frame 35 is 000000000000000000, not 000009…), since #1138 fixed HTTP.make.

Probe (options-transport.pcap, frames 35–44, one per type): before, 4 TCP:HTTP/2 and 6 TCP:Raw (35–38, 40, 44); after, all 10 TCP:HTTP/2. No other generated capture changes. Captures are gitignored and generated, so none is committed.

Test: new tests/protocols/test_option_generator_httpv2_stream_unit.py (unit tier, no capture). Without the fix: 14 failed, 4 passed, 27 subtests passed; with it: 5 passed, 40 subtests passed. No existing test pinned the Raw decoding. Also passing: test_option_roundtrip_unit, test_option_coverage_runtime, test_fields_copy_dispatch_runtime, test_tcp_http_dispatch_unit, test_option_generator_tcp_base_unit, test_generated_pcap_runtime, and tests/project (389 passed, 1 skipped).

[JarryShaw]

Cross-review verdict on dcd727ac5: GOOD TO GO (ran on Sonnet; author Opus)

• Root cause confirmed: on main, frames 35–44 all sit on stream 0 with a payload-only Length. _test_http2_frame accepts types 0/1/2/3/5/9 only on a non-zero stream, so exactly the six Raw frames were these types (35, 36, 37, 38, 40, 44).
• Stream assignments match RFC 9113 §6.1–6.10. PUSH_PROMISE on client stream 1 promising even stream 2 is coherent under §6.6 and §5.1.1.
• Decoding: options-transport.pcap decodes 4/6 HTTP/2 vs Raw before and 10/0 after. The other four options-* captures decode identically.
• The new test is independent and fails without the fix: it carries its own literal RFC table, and gives 14 failed with options.py reverted.
• Composes cleanly with fix(http): HTTP.from_data raises 'invalid HTTP version: 0' #1162: merged together on main, this PR's test, test_option_roundtrip_unit and fix(http): HTTP.from_data raises 'invalid HTTP version: 0' #1162's test all pass.
• Follow-ups, not blocking: the sample WINDOW_UPDATE carries a zero increment, which RFC 9113 §6.9 treats as an error (this predates the PR). Also, the test docstring's example failure text would read False != True, not 0 != 1.

[github-actions]

Coverage: 88.87% (unit tier, Python 3.14, dcd727ac5, Unit Tests run success)

Package	Statements	Missed	Branches	Partial	Cover
pcapkit (top level)	104	4	20	4	93.55%
pcapkit/const	18757	1035	2342	846	90.23%
pcapkit/corekit	1878	91	580	22	94.34%
pcapkit/dumpkit	142	0	42	0	100.00%
pcapkit/foundation	2422	104	842	40	94.24%
pcapkit/interface	112	7	40	5	92.11%
pcapkit/protocols	15688	188	3960	163	98.18%
pcapkit/toolkit	487	65	144	1	85.42%
pcapkit/utilities	429	4	122	4	98.55%
pcapkit/vendor	4409	2359	1006	158	42.84%

Per-file detail: the coverage-html artifact of this run.