Correlated variables via the simstudy backend (gated on seed ADR)

[DougManuel]

Deferred-register item in the post-v0.4.0 development plan (development/post-v040-development-plan.md). Do not start before the seed-contract ADR lands — this adds a third RNG consumer.

Scope guards (why deferred)

• cran-comments.md and design-philosophy-v04.qmd:257-259 explicitly disclaim statistically realistic joint distributions and synthetic-data-for-inference. Correlation is the feature that erodes that boundary; each increment of realism invites misuse.
• The honest testing case is narrow: derived variables and joint filters need non-degenerate joint mass — crude correlation suffices; fidelity is a non-goal.

Shape when it happens

• simstudy-backend-only (genCorData/copulas exist there; do not implement correlation natively — the hybrid-backend ADR chose simstudy precisely to avoid rebuilding a distribution/correlation engine).
• Opt-in; documented loudly as plausibility-not-fidelity.
• Open design question from the ADR: merge strategy between correlated groups and ordinary variables (column ordering, diagnostics attribution, reproducibility across the two generators).

Trigger to activate

A concrete downstream need (e.g., a derived-variable test that cannot get coverage from independent marginals), plus seed ADR merged.