test(project): evaluate the release gates' if: expressions (#962)

[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 — 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?

• 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

Closes #962. .github/workflows/ is untouched — the workflow is the fixture.

What is now established. tag/pypi/conda are evaluated over 17/17/21
scenario rows: all four values of needs.version_check.result (twice — outputs
present, which is what catches the clause being deleted, and outputs absent, the
skipped shape), all four of each publishing predecessor's, a cancelled run, and
all four corners of startsWith(ref_name, 'v') || <evidence> == 'false' with the
other three evidence outputs inverted so a gate reading the wrong one fails.
github and unit-tests get their own tables. One axis per row, everything else
held at a running value — a row holding two clauses false proves less than it looks.

Mutations. 14 applied to a scratch copy of each real condition, asserting
which layer rejects each. Caught by both layers: &&→|| (either side),
== 'success'→!= 'failure', || == 'skipped' added, contains(fromJSON(…)),
the conjunct dropped, the predecessor pair weakened to != 'failure', || == 'skipped'
removed, the evidence output swapped. Evaluator only: true || prefix (#961's
recorded residual), !cancelled()→always(), startsWith prefix emptied. Text
only, and correctly so: reversed operands — same truth table, so the evaluator has
nothing to say and the retained text assertion is what notices. Nothing survives both.

Residuals. The implicit success() Actions prepends to github's if:, and the
cascade-skip, are runner behaviour and stay pinned textually. Case-only mutations are
invisible to the evaluator by construction (Actions compares case-insensitively).

35 → 58 tests, 188 subtests; pytest -q and plain unittest agree.

[JarryShaw]

GOOD TO GO at e403c9166 — sonnet cross-review, first round. Different model
from the author (opus).

Re-derived on my side rather than relayed: 58 tests pass under plain unittest
in the PR's own tree (pcapkit.__file__ asserted into that worktree), 35 → 58 is
23 new, and the only 8 deleted lines are the stale "Knowingly not closed"
docstring paragraph — the hole this PR closes. No test removed or weakened.

It found a fifteenth mutation that survives both layers, and I confirmed it.
Rewriting an evidence comparison from == 'false' to != 'true' yields zero
violations from the evaluator and zero from the text mirror, on all three jobs
(baseline condition clean as a control). The text layer only checks the evidence
name appears; every modelled row holds the outputs at literal 'true'/'false',
so the two spellings never diverge on a modelled row.

One correction to the review's own reasoning: it called this unreachable because
version_check's steps always emit a literal. True of the two curl checks, which
default complete=false then set true. Not true of PCAPKIT_TAG_EXISTS and
PCAPKIT_CONDA_TAG_EXISTS, which come from mukunku/tag-exists-action@v1.7.0 —
a third-party output this repo does not control, and tag's gate reads one. So
the gap is narrower than a defect but realer than unreachable.

Not blocking: it is the same character as the four exclusions the module already
discloses, and worth less than the 23 tests it would hold up. Filed as a
follow-up.