Self-contain vignettes/examples on bundled data; label interaction scale

[ehsanx]

Removes the last dependency on the Suggests-only NHANES package and makes the additive-interaction scale explicit in the output. R CMD check --as-cran --run-donttest remains clean (0 errors, 0 warnings, 1 expected NOTE).

Vignettes (now build without NHANES installed):

• Convert v1, v2, v3, v5 off NHANESraw onto the bundled nhanes_mortality cohort (hypertension / insulin / caffeine analyses); fix the malformed inline setup chunks in v1 and v2.
• v4 already used bundled data; guard its mitools (Suggests) use with eval = requireNamespace("mitools", ...) and qualify mitools::imputationList.

Interaction interpretation safety:

• addintlist() now returns a Scale column ("OR (logistic)", "HR (Cox)", "RR (log-binomial)", "RR (Poisson)", or "ratio") inferred from the model, so the effect-measure scale of RERI/AP/S is visible in the output itself.
• Document the rare-outcome RR-vs-OR caveat in addint() and addintlist() @details, citing VanderWeele and Knol (2014).

Self-containment:

• Convert the remaining NHANES examples (addintlist, jointeffects, plotint, svytable1, svydiag) to bundled data; svytable1/svydiag stay runnable (plain) examples, verified fast and lonely-PSU-safe.
• Drop NHANES from Suggests (no longer referenced anywhere).

Tests: 69 passing (adds Scale-column and svykmplot se = TRUE coverage).