v0.9.0

New features

• set_higher_is() and extract_higher_is() store and retrieve a
  direction-of-improvement attribute ("better" or "worse") for survey
  variables. The attribute is used by get_diffs(show_favorability = TRUE)
  to classify differences as favorable or backlash. Supports all three
  calling conventions (named ... args, named vector,
  variable = + direction =), data frames, and tidy-select.

• set_reverse_coded() and extract_reverse_coded() store and retrieve a
  reverse-coded flag for survey variables. Supports tidy-select ... (bare
  names and selection helpers) and the variable = character-vector
  interface. Setting reverse_coded = FALSE removes the flag. Works on both
  survey design objects and plain data frames.

• get_diffs() gains alpha and show_favorability arguments.
  show_favorability = TRUE appends favorable and backlash logical
  columns to the result. A difference is classified as favorable when it is
  statistically significant (p < alpha, default 0.05) and in the
  direction indicated by higher_is metadata set via set_higher_is().
  When no higher_is metadata is set, both columns are all FALSE. The
  adjusted p-value (when pval_adj is supplied) is used for classification.

• as_survey_nonprob() gains replicate-weight (bootstrap) support via six
  new arguments: repweights, type, scale, rscales, mse, and
  reference_sample. When repweights is supplied, the constructor resolves
  and validates the replicate weight columns and stores them in @variables.
  A reference_sample property is also added to the survey_nonprob class
  for provenance tracking.

• All get_*() estimation functions now dispatch to the replicate-weight
  variance engine when a survey_nonprob object has repweights set.
  When no repweights are supplied, a warning is emitted and the SRS
  approximation is used as a fallback. Degrees of freedom for
  survey_nonprob estimates always use the normal approximation (Inf).

• print() for survey_nonprob now shows bootstrap replicate information
  (replicate count, type, and scale) when repweights are set.