Escape unlinked cross-references in HTML snippets (#1833)

## Description

`RDoc::Markup::ToHtmlSnippet` doesn't link cross-references, but its
`handle_regexp_CROSSREF` returned the matched text unchanged.
`apply_regexp_handling` treats a handler's return value as
already-converted HTML, so a CROSSREF match such as `#<` in
`[#<Encoding:ISO-8859-1>, #<Encoding:UTF-8>]` was emitted with a raw
`<`. The resulting search snippets (Aliki search results, Darkfish
`search_index.js`) contained `#<Encoding:UTF-8&gt;`, which the browser
parses as a tag.

The handler now escapes the text with `convert_string`, the same way
`ToHtml#handle_regexp_SUPPRESSED_CROSSREF` does. On current master the
regular page output (`ToHtmlCrossref`) already escapes this case, so the
snippet formatter was the remaining path; I checked by generating aliki
and darkfish docs for the example from the issue, before and after.

Closes #1743

Author: anyingiit

lib/rdoc/markup/to_html_snippet.rb

@@ -128,10 +128,11 @@ def start_accepting
       end
 
       ##
-      # Removes escaping from the cross-references in +target+
+      # Removes escaping from the cross-references in +target+ and escapes HTML
+      # characters, since snippets don't link cross-references.
 
       def handle_regexp_CROSSREF(text)
-        text.sub(/\A\\/, '')
+        convert_string(text.delete_prefix('\\'))
       end
 
       ##

test/rdoc/markup/to_html_snippet_test.rb

@@ -650,6 +650,13 @@ def test_convert_TIDYLINK_rdoc_label
     assert_equal 3, @to.characters
   end
 
+  def test_convert_CROSSREF_escapes_html
+    result = @to.convert 'See [#<Encoding:ISO-8859-1>, #<Encoding:UTF-8>] and Foo::<bar>'
+
+    expected = "<p>See [#&lt;Encoding:ISO-8859-1&gt;, #&lt;Encoding:UTF-8&gt;] and Foo::&lt;bar&gt;\n"
+    assert_equal expected, result
+  end
+
   def test_handle_regexp_HYPERLINK_link
     target = 'link:README.txt'