add path based redirects

up to now, those were managed outside of this repo, which as
unsurprisingly a real hassle to deal with if one wanted to prevent URLs
from breaking when moving pages around. this change removes a large part
of the friction involved in moving content in the Nix manual.

possible next steps for further automation:
- check for content that moved and warn if it's not reachable from
  links that were valid prior to a change
- create redirect rules automatically based on this information

(cherry picked from commit 2b7016cc56)

doc/manual/redirects.js

@@ -1,7 +1,9 @@
-// redirect rules for anchors ensure backwards compatibility of URLs.
-// this must be done on the client side, as web servers do not see the anchor part of the URL.
+// redirect rules for URL fragments (client-side) to prevent link rot.
+// this must be done on the client side, as web servers do not see the fragment part of the URL.
+// it will only work with JavaScript enabled in the browser, but this is the best we can do here.
+// see ./_redirects for path redirects (client-side)
 
-// redirections are declared as follows:
+// redirects are declared as follows:
 // each entry has as its key a path matching the requested URL path, relative to the mdBook document root.
 //
 //     IMPORTANT: it must specify the full path with file name and suffix