tf 0.4.0

New features

tf_register() & co implement registration methods (SRVF, affine, landmark, continuous-criterion) for aligning functions and analyzing phase variability. Returns a new tf_registration class with accessors tf_aligned(), tf_inv_warps(), tf_template().
tf_depth() now implements five functional depth methods: MBD (default), MHI, FM (Fraiman-Muniz), FSD (Functional Spatial Depth), and RPD (Regularized Projection Depth). MHI ranks functions from lowest (0) to highest (1); all other depths are scaled from 0 (most extreme) to 1 (most central). MBD is now scaled to [0, 1] instead of [0, 0.5].
tf_split() / tf_combine() for splitting functions into sub-domain fragments and re-combining them
tf_rgp() gains a "brownian_bridge" covariance option and can generate irregular data via the new irreg argument
tfb_spline() now converts fda::fd and fda::fdSmooth objects directly
mgcv-style Fourier basis constructor for use in tfb_spline()
sparklines in print() / format() for all tf subtypes
tf_where() now defaults to arg = tf_arg(x), consistent with other functions
as.matrix() for irregular tfd now always interpolates to a common grid

tf_derive() for tfd objects now uses second-order accurate finite differences, preserving the original domain and grid (#59)
major overhaul of NA handling: NA entries are now consistently represented as NULL internally; fixes for tf_arg(), tf_evaluate(), tf_smooth(), tfd() conversion, is.na(), printing, and arithmetic with NA entries
fix summary.tf() returning wrong central range
fix tfd domain enforcement for length-1 data (#139)
fix printing for empty tfb objects and all-NA vectors

tf_derive and tf_integrate can now handle tfb with non-identity link functions by falling back to tfd-calculus methods (and returning tfd objects).
tfd_irreg arithmetic now operates on intersection of arg-values instead of failing when arg-values are not identical
tfb arithmetic fix
dropped glue dependency in favour of cli / sprintf; replaced stopifnot() with informative cli::cli_abort() throughout
use vctrs::vec_arith() for group generics, vctrs utils in [.tf
various internal performance improvements
use air formatter for all R code

tf_rebase behaves more consistently on (irregular) data with different lengths
avoid CRAN issues by streamlining tfb_fpc example and skipping tf_rebase tests