New datasets, traceCI(), standardized coefficients, and 3D heplots

Version 1.8.5

• Added the LearnDis dataset (Tabachnick & Fidell, 2013, Table 7.1): a small
  3 (Disability: Mild/Moderate/Severe) x 2 (Treatment/Control) factorial giving
  reading (WRAT_R) and arithmetic (WRAT_A) achievement scores and IQ, for
  worked MANOVA/MANCOVA examples and, in particular, a citable, hand-checkable
  example for the Roy-Bargmann stepdown analysis under development. Its
  Disability factor is ordered (Mild < Moderate < Severe).

• Added the ReadingDisability dataset: 571 children classified into four
  reading-achievement groups (Severe/Mild/Average/Superior, an ordered
  factor) on six cognitive/achievement measures (PPVT, RD, EF, VF,
  VMI, SIM). Reconstructed by simulation from the group means, SDs, and
  pooled within-cells correlations in Bray & Maxwell (1985, Multivariate
  Analysis of Variance, Table 2.3).

• Added traceCI(): analytic (Bai & Silverstein, 2004) confidence intervals
  for the trace (sum of eigenvalues) of one or more covariance matrices,
  complementing logdetCI() and eigstatCI().

• Added stdmodel() and stdcoef() for standardized ("beta") coefficients on
  lm/mlm objects: responses and numeric predictors are standardized, factor
  predictors are left raw. coefplot.mlm() gains a std = TRUE argument using
  this to plot standardized-coefficient confidence ellipses.

• Added a live heplot3d() example to the HE_manova vignette's MockJury
  section (ratings on phyattr, independent, and sophisticated), shown as
  a rotatable, zoomable 3D widget.

Maintenance

• Clarified the coefplot.mlm() documentation to explain how it differs from
  car::confidenceEllipse()'s mlm method: coefplot() fixes a pair of
  responses as plot axes and overlays one ellipse per predictor, while
  confidenceEllipse() fixes a pair of coefficients and shows their joint
  confidence region.

• Added an explicit rgl::rglwidget() call to heplot3d()'s example, so its
  pkgdown reference page shows an interactive 3D plot instead of nothing.
  rgl's automatic pkgdown widget support only fires for a visibly-returned
  object of class rglId/rglOpen3d; heplot3d() returns a different class
  invisibly, so nothing triggered it without this explicit call.

• Calling rgl::rglwidget() directly in a bookdown-based, as_is vignette
  (like HE_manova's new example above) renders fine in a standalone vignette
  build but shows as a blank box on the pkgdown site: pkgdown replaces the
  <head> of the article page when assembling it, silently dropping the
  <script> tags the widget needs. The fix -- pre-rendering the widget to a
  self-contained file and embedding it with an <iframe> instead -- is
  documented in heplot3d()'s @details for reuse in other vignettes.

• Fixed math rendering on the pkgdown site for all four as_is vignettes
  (HE_manova, HE_mmra, Robust, datasets): _pkgdown.yml never
  configured a math renderer, so $...$ expressions showed as literal text
  instead of typeset equations. Configured KaTeX (matching the working setup
  already used in the matlib package) after finding that MathJax's CDN
  bundle doesn't support \boldsymbol, needed for bold Greek letters like
  \boldsymbol{\beta} (\mathbf{} only covers Roman symbols, not Greek).

• Bare/dollar-math H/E references to the hypothesis and error matrices (in
  heplot(), heplot1d(), heplot3d(), pairs.mlm(), etasq(), and the
  package overview) are now \eqn{\mathbf{H}}{H}/\eqn{\mathbf{E}}{E} -- the
  earlier $\mathbf{H}$ form isn't valid Rd and rendered as mangled literal
  text in plain ?fun help.

• Modernized roxygen documentation across R/: legacy Rd macros (\code{},
  \link{}/\link[pkg]{}, \pkg{}, \emph{}) converted to markdown syntax
  in 44 files, fixing two pre-existing broken links (missing-backslash typos)
  and two stale rgl link targets found along the way.

• Reorganized vignette assets: images and figures moved from vignettes/fig/
  to vignettes/images/, and four orphaned, unreferenced 2017-era images
  removed.

• Dropped the hard Depends: broom for glance.mlm(); now Imports: generics
  (the lightweight package defining the glance generic, which broom itself
  depends on) instead, with broom moved to Suggests. glance() still works
  the same as before with just library(heplots).