fix(tests): make two tests hold on free-threaded CPython 3.14t (#1731)

[JarryShaw]

Please follow the guide below

• You will be asked some questions, please read them carefully and answer honestly

• Put an x into all the boxes [ ] relevant to your pull request (like that [x])

• Use Preview tab to see how your pull request will actually look like

• 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. Narrow runs only, listed below.

• 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?

Tick the commit type your subject line carries.

• 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
• release — bumps the version or rolls up a distribution
• chore — anything else

---

Description of your pull request and other information

Closes #1731. Two tests failed on free-threaded CPython 3.14t. Each came from a wrong assumption in the test, not a pcapkit defect, so no library code changes.

• test_hold_is_local_to_its_context: under context-aware warnings, catch_warnings records through a ContextVar that an empty Context lacks. Free-threaded builds turn that on by default (CPython docs: -X context_aware_warnings). The failure follows the flag both ways: it fails on regular 3.14 with =1 and passes on 3.14t with =0. The test now warns in a context copied before the block, and in an empty context that records for itself.
• test_the_captured_len_vector_parses_under_a_memory_cap: this one is not about warnings, since it fails on 3.14t with the flag off. mimalloc reserves 1 GiB at start-up (Objects/mimalloc/options.c:86-87), so the 512 MiB RLIMIT_AS was used up before the parse began. The cap is now 512 MiB above what is already mapped. The child also runs under PYTHONMALLOC=mimalloc where the build has it, so a GIL leg covers this case too.

Tests:

• Both files pass on 3.14t, and on 3.14 with the flag at 0 and at 1, plus on 3.11 and 3.8.
• On main both fail on 3.14t.
• An unbounded-padding mutant and a global-hold mutant each still fail the fixed tests.

[JarryShaw]

Verdict: GOOD TO GO at ba67ef70e. A Sonnet cross-review checked it on its own free-threaded 3.14t (author: Opus).

• Two separate causes:
  • The hold test's failure follows -X context_aware_warnings both ways.
  • The memory-cap test fails on 3.14t even with that flag off, because mimalloc's 1 GiB start-up reservation uses up the 512 MiB RLIMIT_AS. Checked against CPython's Objects/mimalloc/options.c.
• Not weakened: both tests still catch their mutants (unbounded padding, a global hold). No library code changes.
• Interpreters: both pass on 3.14t, on 3.14 with the flag at 0 and at 1, and on 3.11 and 3.8.

[github-actions]

Coverage: 89.65% (unit tier, Python 3.15, ba67ef70e, Unit Tests run failure)

Package	Statements	Missed	Branches	Partial	Cover
pcapkit (top level)	104	3	20	3	95.16%
pcapkit/const	18860	954	2368	868	91.06%
pcapkit/corekit	2384	67	764	35	96.51%
pcapkit/dumpkit	275	3	100	3	98.40%
pcapkit/foundation	2885	144	1010	34	94.20%
pcapkit/interface	138	7	52	5	93.68%
pcapkit/protocols	17174	226	4590	190	98.07%
pcapkit/toolkit	1015	173	372	10	80.89%
pcapkit/utilities	508	4	154	4	98.79%
pcapkit/vendor	4417	2342	1010	157	43.38%

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