This reverts commit 50cdf076.

The default is for hyphens to separate tokens in search terms: `gem-based` is equivalent to `gem based`.

This adds `search_tokenizer_separator` as a site configuation option, to support search for hyphenated words.

(cherry picked from commit 99371e971406090a2dba7b9b69f514a8f87afce3)

(cherry picked from commit 468264941d08b00252a4f6f081bc2dff5920bffe)

(cherry picked from commit 9f3f38acf4e1ac40e3fb2b2f86018d4cba28dd07)

(cherry picked from commit 86ff2dac05412a6ff994968c2be6122a7fcf85f9)

(cherry picked from commit d39b7b3041d1bdb9f80ab7f5878e64a3b3b79b5d)

(cherry picked from commit f25d76730b175fb680aa2770fb20a67fd858ffbb)

(cherry picked from commit db50f282bab1e98397cc7434a811b8b7da2fe6dc)

(cherry picked from commit 2b153ed074811e735aaa58d57a3fcdf7838e17f5)

This commit makes the dark mode preview feel a little more interactive
and it also pulls some duplicated code into one file so that people like
me aren't stumped when they are looking at the wrong page!

If you exclude the last page in the site.html_pages through exclude_true you end up violating the assumption in the for loop that the last page won't have a trailing comma. This breaks the json data that is served to the front end.

I removed the forloop.last check and instead assign a temporary variable to check for the first non-excluded page, and then add the comma before the next non-excluded entry.

Because `page.title` (and `page.url`) is already escaped and page titles served by lunrjs do not need escaping, ampersands need to be unescaped to display properly in search results.

and mention `search_exclude`