feat: accept svyrep.design in the survey test and CI functions

[szimmer]

What changes are proposed in this pull request?

• ard_survey_svychisq(), ard_survey_svyttest(), ard_survey_svyranktest(), ard_categorical_ci() and ard_continuous_ci() now accept replicate-weight survey designs (svyrep.design). These functions previously rejected such designs with a class check, although the survey functions they wrap already support them. This is what gtsummary::add_p() and gtsummary::add_ci() require: because add_p() traps errors per variable, the rejection surfaced to users as a silently empty p-value column rather than an error. (Feature Request: Support svyrep.design (replicate-weight) survey objects #355, @szimmer)

Provide more detail here as needed.

This is the second of the two blockers described in #355, and follows #356, which added the missing svyrep.design methods for the six summary ard_* generics.

The guards were stricter than the code they wrap. Each of the five functions carried check_class(data, "survey.design"), but the underlying survey functions handle replicate designs already — svychisq(), svyranktest() and svyquantile() have dedicated svyrep.design methods, and svyttest()/svyciprop() are not generic and are design-agnostic:

svychisq(~stype + both, des)  #> p = 0.06135439
svychisq(~stype + both, rep)  #> p = 0.07992227

ard_categorical_ci() and ard_continuous_ci() also needed svyrep.design methods registered, since they are S3 methods rather than plain functions — relaxing the guard alone would not route a replicate design to them. I placed those two methods alongside their survey.design counterparts rather than in R/svyrep.design.R, so this branch does not conflict with #356. Happy to consolidate them into that file once both are merged, if you'd prefer them together.

The user-visible effect. gtsummary::add_p() traps errors per variable, so the class guard did not reach users as a message — it produced a table with an empty p-value column. With both PRs in place:

	before	after
add_p(), replicate design	NA	0.07992227
svychisq() called directly	0.07992227	matches
add_ci(), replicate design	error	588, 701

Tests (tests/testthat/test-svyrep.design_tests_ci.R) assert for each function that the result matches survey computed directly on the replicate design, and that it differs from the same computation on the linearized design. A silent fallback to linearization would therefore fail rather than pass. The class guards are also tested to still reject non-survey input.

Not included. ard_emmeans_contrast() and ard_emmeans_emmeans() carry the same guard and are left for a follow-up — I have not yet tested emmeans against replicate designs, and did not want that holding up the five functions I can evidence. gtsummary::add_difference() will therefore still fail on replicate designs after this merges.

Reference GitHub issue associated with pull request.

Addresses #355. Together with #356 this covers both blockers described there, apart from the ard_emmeans_* pair noted above.

---

Pre-review Checklist (if item does not apply, mark is as complete)

• All GitHub Action workflows pass with a ✅
• PR branch has pulled the most recent updates from master branch: usethis::pr_merge_main()
• If a bug was fixed, a unit test was added.
• If a new ard_*() function was added, it passes the ARD structural checks from cards::check_ard_structure().
• If a new ard_*() function was added, set_cli_abort_call() has been set.
• If a new ard_*() function was added and it depends on another package (such as, broom), is_pkg_installed("broom") has been set in the function call and the following added to the roxygen comments: @examplesIf do.call(asNamespace("cardx")$is_pkg_installed, list(pkg = "broom""))
• Code coverage is suitable for any new functions/features (generally, 100% coverage for new code): devtools::test_coverage()

Notes on the checklist:

• No new exported functions are added; the two new S3 methods delegate to their survey.design counterparts, so set_cli_abort_call() and the is_pkg_installed() checks are inherited.
• covr hits every line this PR changes (7/7). Per file: ard_continuous_ci.survey.design.R 100%, ard_categorical_ci.survey.design.R 96.67%, ard_survey_svyttest.R 93.75%, ard_survey_svyranktest.R 93.33%, ard_survey_svychisq.R 91.67% — the shortfalls are pre-existing uncovered lines this PR does not touch. Package coverage is 96.73%, unchanged from main.
• lintr reports no findings on any changed line.