feat: support svyrep.design (replicate-weight) survey objects

[szimmer]

What changes are proposed in this pull request?

• Added svyrep.design methods for ard_attributes(), ard_total_n(), ard_missing(), ard_tabulate(), ard_tabulate_value() and ard_summary(), so these functions now accept replicate-weight survey designs created with survey::svrepdesign() or survey::as.svrepdesign(). Because svyrep.design is a sibling class of survey.design rather than a subclass, such designs previously failed S3 dispatch. Statistics are computed by the same survey functions as for survey.design, all of which already support replicate designs, so variance estimates correctly reflect the replicate weights. The survey test and confidence interval functions (e.g. ard_survey_svychisq(), ard_categorical_ci(), ard_continuous_ci()) do not yet accept replicate designs. (Feature Request: Support svyrep.design (replicate-weight) survey objects #355, @szimmer)

Provide more detail here as needed.

svyrep.design is a sibling of survey.design, not a subclass:

class(survey::svydesign(...))      #> "survey.design2" "survey.design"
class(survey::as.svrepdesign(...)) #> "svyrep.design"
inherits(rep, "survey.design")     #> FALSE

so replicate designs were rejected at dispatch, before reaching any statistical code.

No new statistical code is introduced. Each new method is a thin wrapper that forwards to its survey.design counterpart, which reaches survey through .compute_svy_stat() and .svytable_counts():

survey function called	replicate support
svymean, svytotal, svyvar, svyquantile, svytable	dedicated svyrep.design methods
svyby	handled by svyby.default
SE, deff	class-agnostic extractors

Scope. This covers the first of the two blockers described in #355 — the missing S3 methods. The seven functions with hard check_class(data, "survey.design") guards are untouched here, so gtsummary::add_p() and add_ci() will still fail on replicate designs after this merges. I'd like to follow up with a second PR relaxing five of them (ard_survey_svychisq(), ard_survey_svyttest(), ard_survey_svyranktest(), ard_categorical_ci(), ard_continuous_ci()), each of which wraps a survey function that already handles replicate designs, and a third for ard_emmeans_contrast()/ard_emmeans_emmeans(), which I have not yet tested against replicate designs. Splitting it this way keeps each PR to changes I can evidence. Happy to combine them if you'd rather review one.

Tests (tests/testthat/test-svyrep.design.R) cover:

• dispatch of all six generics on a svyrep.design
• cards::check_ard_structure() on the results
• weighted point estimates (n, N, p, median) agreeing exactly with the equivalent linearized design, since the sampling weights are identical
• replicate variance genuinely being used — the p.std.error values differ from the linearized design and match survey::SE(survey::svymean(...)) on the same design, so this is not a silent fallback to linearization
• every replicate type: JK1, JKn, bootstrap, subbootstrap, Fay, and BRR built directly with svrepdesign()
• by= tabulation

For reference, on apiclus1:

stype	linearized SE	replicate SE
E	0.04633655	0.05144131
H	0.02681022	0.02777990
M	0.02964905	0.03317430

Motivation. Replicate weights are the standard variance-estimation method shipped with most large public-use survey files — NHANES, NHIS, BRFSS, ACS and the NCES surveys, and anything using BRR or jackknife. This is also the blocker for gtsummary::tbl_svysummary() supporting replicate designs (ddsjoberg/gtsummary#1441, #1535).

Reference GitHub issue associated with pull request.

Addresses #355 (the first of the two blockers described there; please leave the issue open for the check_class() guards).

---

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:

• covr reports 100% for R/svyrep.design.R (6/6 lines, exercised by the new tests); package coverage is 96.73%.
• set_cli_abort_call() and is_pkg_installed("survey") are inherited from the survey.design functions the new methods delegate to.
• The methods share help topics with their survey.design counterparts via @rdname; each topic gained an @examplesIf block showing a replicate design.