* TODO [Enhancement] Support lack of Imagemagick type in upcoming Emacs release

Current strategy for viewing images:

- Insert alt text in textual Emacs
- If Imagemagick is available, insert scaled image
- Let =create-image= figure out an available image type
- If that fails insert alt text instead of image

The upcoming Emacs release disables Imagemagick support by default for
security reasons and supports native image transformations.  It's
possible to test for the latter with the following snippet:

#+BEGIN_SRC emacs-lisp
(and (fboundp 'image-transforms-p) (memq 'scale (image-transforms-p)))
#+END_SRC

The above strategy should be reworked to scale the image if possible
(preferring native, then Imagemagick), then fall back to native
formats, then plain text.

* TODO [Enhancement] Handle displays not capable of displaying images

Currently this is tested with =display-graphic-p= instead of
=display-images-p=.  Another thing to consider is whether an image
created with =(insert-image ... ALT)= works better than inserting
text.

* TODO [Bug] SVGs containing images are not rendered correctly

This seems to mostly happen with librsvg as it fails to resolve these
relative to the current directory.  Might require Emacs patches.

An alternative "solution" is to edit SVGs to not contain relative
paths after unzipping...

* TODO [Enhancement] Remember document position for navigation with t/n/p

Much like with remembering the last viewed position per file, you
could remember it per document.  It doesn't make sense for navigation
with =SPC= and =S-SPC= to do this though.

* TODO [Feature] Support FB2 format

This basically amounts to a full rewrite using generic functions.  It
doesn't help the format is used for Russian literature mostly and is
documented in terms of a few XML files.

http://gribuser.ru/xml/fictionbook/index.html.en

* TODO [Feature] Info-like continuous search

There are a few approaches to this, none of them strike me as good
enough to pursue:

- Using grep to do fuzzy matching, then jumping to the "right"
  location.  Tricky because there is no correspondance between
  rendered buffer and HTML source.  Rejected on emacs-devel:
  https://lists.gnu.org/archive/html/emacs-devel/2020-02/msg00096.html
- Using =dom-texts= on the HTML parse trees, then jumping to the
  "right" location using fuzzy matching of the text.
- Using isearch in one document, then loading up the adjacent one if
  needed.  Slow unless caching is used and implementing that is
  another can of worms.