inlabru 2.15.0

New features

• New predictor evaluation and linearisation implementation (2.14.1.9008)
• Add tidy(), glance(), and augment() methods for bru objects, to
  support the broom package tidying framework. Thanks to Novica Nakov
  (2.14.1.9005)
• Add observation family options nzbinomial, nznbinomial, nzcenpoisson,
  and nzbetabinomial for non-zero truncated models, acting like nzpoisson.
  Will automatically switch to INLA implementations if available, otherwise
  uses zeroinflated*0 with the zero-probability fixed to 2e-9.
  (2.14.1.9007)
• Improved bru_used variable detection, supporting pronoun and container
  object access detection. (2.14.1.9010)
• Add log argument to bru_timings_plot() to allow logarithmic scaling of
  the time axis (2.14.1.9014)

Bug fixes

• Fix issue #298, where weights input to copy = ... components was not
  consistently passed to the component mapper construction. This could be
  partly worked around by specifying weights = 1 in the source component, but
  only for estimation, so post-estimation computations would still ignore the
  weights.
  Now enforces that the same inputs are present in both the source and copy
  definitions, ensuring a consistent mapper construction, and post-estimation
  consistency. (2.14.1.9002)
• (Re)enable non-integer point weights for family = "cp" models, by using
  family = "xpoisson" (2.14.1.9003)
• Fix layer indexing in bru_fill_missing() for SpatRaster data that resulted
  in an error for layers beyond the first layer (2.14.1.9006)

Namespace changes

• Remove dependency on plyr. Fixes #301 (2.14.1.9004)
• Move stats from Depends: to Imports: to avoid unnecessary namespace
  clashes (2.14.1.9013)
• Reexport generics tidy(), glance(), augment(), and generate() from
  generics and predict() from stats (2.14.1.9014)