When spawning a new atril instance for cross-document links, the
destination and search parameters from the document were interpolated
directly into the command line without shell quoting. Values containing
spaces or special characters could be split into separate arguments by
the shell parser, potentially being interpreted as unintended flags by
the child process.

Apply shell quoting to page label, named destination, and search string
values before appending them to the command line, consistent with how
other spawn sites in the codebase already handle this.

This is a rather non-invasive change, that will
be easy to remove whenever we go ahead and stop
supporting older systems.

A WebKit processes should only be created when the user actually requests to open a file type requiring WebKit to be processed.

Commit adapted for atril, picked up from:
https://gitlab.gnome.org/GNOME/evince/-/commit/c3de8e75d6d0920478af210ba19a2d94b0734917

Credits to Benjamin Berg <bberg@redhat.com>

  | For large documents the linear search for the first link that is on a
  | certain page is really slow. Because of this scrolling becomes slow
  | whenever the page changes.
  |
  | Replace the linear search with a search in a binary tree populated with
  | the first link on each page and the corresponding GtkTreePath. This way
  | a specialized binary tree lookup can be used to find the closest
  | matching link and select that in the treeview.
  |
  | https://bugzilla.gnome.org/show_bug.cgi?id=779614