Trimming and truncation improvements

[malcolmbarrett]

• add failing tests for refusing a bare continuous conditional mean
• refuse a bare continuous conditional mean in the modified score routes
• add failing tests for modified scores in the data-frame weight route
• keep modified score records in the data-frame weight route and refuse them for a continuous exposure
• add failing tests for trimming a dose model on the density scale
• trim a dose model on the scale of its conditional density
• add failing tests for the dose-aware refit and weights from a density trim
• refit a dose model on the kept rows and build weights from its trim record
• add failing tests for the psw truncation record
• add a truncation record to propensity score weights
• add failing tests for wt_trunc()
• add wt_trunc() to bound extreme weights
• add failing tests for ipw() refusing truncated weights
• refuse truncated weights in ipw()
• add failing tests for the truncation sensitivity grid
• add wt_trunc_sensitivity() to tabulate weights over a grid of bounds
• add failing tests for ps_trunc(method = "adaptive") on binary scores
• add an adaptive bound to ps_trunc() for binary scores
• add a failing test for ps_refit() on a two-level multinom fit
• refit a two-level multinom through its probability prediction
• add failing tests for trimming a binomial gam
• drop the dim attribute of a gam's fitted values
• add failing tests for ps_refit() with a subset expression
• evaluate expressions passed to ps_refit() against the retained data
• refuse a refit whose model does not match the shape of the trimmed scores
• add failing tests for refitting a trim with a first-level focal level
• keep the focal level when refitting a trimmed score
• add failing tests for the ps_trunc(method = "ps") default bounds
• mirror a single supplied bound in ps_trunc(method = "ps")
• add failing tests for combining a single modified vector
• keep the records of a single modified vector through c()
• tidy the trim parameter comparison and the refit data hint
• keep records through c() of a single modified psw only
• validate a lone ps_trunc bound and tighten the refit data hint
• add failing tests for modified scores combined under an explicit prototype
• drop a score record restored against an explicit prototype
• add failing tests for keeping the calibration record on weights
• keep the calibration record through subsetting and combining
• add failing tests for grouped mutate and unique on modified scores
• drop score records silently and keep them consistent through unique()
• add failing tests for unique() on modified score matrices
• keep unique() on score matrices row-wise
• document wt_trunc(), density-scale trimming, and when to use each
• test unique() on one-row and empty score matrices
• add failing tests for reordering weights with position records
• reindex position records when weights are reordered
• speed up record reindexing, dose trimming and wt_trunc()
• keep the non-positional part of a weight record when its positions are dropped
• drop weight record positions when other data is cast onto them
• add failing tests for slicing modified scores at the same length
• drop score positions when a slice cannot place them
• add failing tests for reordering a per-observation stabilization score
• reorder the per-observation stabilization score with the weights
• add failing tests for record settings through weight combines
• compare and keep record settings when weights are combined
• tidy score handling on casts, assignment and is.na<-
• add failing tests for converting modified score matrices to data frames
• keep the record when a modified score matrix becomes a data frame
• add a failing test for refusing a refit of truncated scores
• explain why ps_refit() refuses truncated scores
• add failing tests for assigning weights with different records
• refuse assigning weights whose records differ
• refuse refitting one column of a trimmed score matrix
• explain refused record casts and document assignment of missing values