surreal 0.0.3

New features

• surreal_app() has a demo that runs in a web browser through Shinylive, with
  nothing to install, at https://r-pkg.thecoatlessprofessor.com/surreal/demo/
  (#5).

• surreal_app() gains two tabs. Search plays back the search that makes a
  dataset, and Selection steps through forward selection among decoy
  predictors (#6, #7).

• surreal_decoys() adds predictors of pure noise to a dataset, so the hidden
  image shows only for the model with the real predictors (#6).

• surreal_path() runs forward selection and records every step. Its plot()
  method draws the share each step explained against the criterion's charge
  for a predictor, the criterion and the residual plot at a step (#6, #7).

• surreal_text_points() and surreal_image_points() return the points that
  surreal_text() and surreal_image() hide, as x and y coordinates (#6).

• surreal_trace() records every iteration of the search that makes a
  dataset, and its plot() method draws the image as it stood at an
  iteration. A step below 1 slows the search down (#6).

Minor improvements and bug fixes

• surreal(), surreal_text() and surreal_image() are much faster and use
  far less memory on large images. The data they return for a given seed is
  unchanged (#6).

• surreal_app() download buttons hand over their files when the app runs in
  a web browser (#5).

• surreal_app() shows the generated message in the Source pane, in its
  download and after Undo (#5).

• surreal_app() shows code that runs when the message has a quotation mark
  or a line break in it (#6).

• surreal_image() finds an image drawn in two light tones, such as light
  gray on white (#6).

• surreal_image() reads an image address with a query string, such as
  picture.jpg?raw=1, in its own format (#6).

• surreal_text() gives a clear error for text with no visible characters
  (#6).