tmsens 0.4.0

Fixed the adjusted trimmed means estimator for higher value trimming (side = "HIGH"), which produced incorrect estimates for outcomes not centred near zero
The permutation p-value is now two-sided, so negative treatment effects are detected, and uses the (b+1)/(n_perm+1) estimator so it is never exactly zero
Fixed an off-by-one in higher value trimming which removed one more observation per group than the specified trimming fraction
Fixed print.tm_bias() erroring when the dropout spread was left unspecified
tm() and tm_bias() now work with any treatment variable name, not just TR
side now defaults to "LOW" and invalid values give a clear error
tm_bias() gains the documented trF default
tm_bias() no longer errors when a treatment group has no dropout
summary.tm() output now shows dropout and post-trimming sample sizes for both treatment groups
tm() no longer perturbs the random number generator state beyond the draws needed for the permutation test
The trimming fraction and treatment variable are now validated up front with clear error messages
Treatment groups are defined by factor level order, fixing numeric group codes such as 2 and 10 and respecting user-set reference levels
The permutation test is around 9 times faster
tm_bias objects gain a groups component giving the treatment and comparator group values
Corrected the author name spelling and publication year in the package citation

tmsens 0.3.1