[v0.6] Table 1 bootstrap adapter (strict-schema v1)

[DougManuel]

Part of the post-v0.4.0 development plan (development/post-v040-development-plan.md, Phase 3, Task 9).

Context

The third input adapter anticipated by the v0.4 architecture (design-philosophy-v04.qmd "What is deliberately deferred"; ADR "Open Follow-Up Decisions"): take published descriptive statistics (a "Table 1") and produce a mock_spec.

Why it matters: teaching and pipeline testing from published statistics, without raw-data access — students and collaborators (including LMIC settings without data-sharing agreements) can build and test against realistic marginals. This is the most distinctive item on the roadmap.

v1 scope — strict input schema only

A tidy data frame in, mock_spec out:

column	type	notes
variable	chr	variable name
type	chr	"categorical" / "continuous"
level	chr	categorical only; one row per level
prop	dbl	categorical only
mean, sd	dbl	continuous only (maps to existing normal support)
min, max	dbl	continuous range

Constructor: mock_spec_from_table1(table1_df) — validation errors name the offending row/column.

Explicit v1 exclusions

• No median/IQR → distribution conversion (later, with a documented heuristic).
• No stratified/grouped panels.
• No parsing of documents, PDFs, or free text — that is a separate tool (possibly an LLM-assisted workflow outside the package).
• Scope guard: outputs match marginals for testing/teaching plausibility. This is not synthetic data for inference or release — same boundary as cran-comments.md and the design philosophy. Vignette must state this.

Acceptance

• Adapter + validation + tests; round-trip vignette (paper table → mock data → summary table comparison).