Preserve encoding on truncate_bytes (rails#51313)

String.new with no arguments returns the empty string with ASCII-8BIT
encoding. Then, depending on each grapheme cluster of the string and
on the omission string, the resulting string might keep the ASCII-8BIT
encoding. With this change, we preserve the encoding of the original
string instead.

Note that String.new accepts an `encoding` keyword argument, like
```
String.new(encoding: Encoding::UTF_8)
```
However, instead of using that, we rely on `force_encoding` to set the
original encoding. This is so that String subclasses don't need to
preserve this keyword argument. For example, SafeBuffer doesn't.
Thanks to @jeremy for catching this!

Author: rosa

activesupport/lib/active_support/core_ext/string/filters.rb

@@ -109,7 +109,7 @@ def truncate_bytes(truncate_to, omission: "…")
     when omission.bytesize == truncate_to
       omission.dup
     else
-      self.class.new.tap do |cut|
+      self.class.new.force_encoding(encoding).tap do |cut|
         cut_at = truncate_to - omission.bytesize
 
         each_grapheme_cluster do |grapheme|

activesupport/test/core_ext/string_ext_test.rb

@@ -376,6 +376,15 @@ def test_truncates_bytes_preserves_grapheme_clusters
     assert_equal "", "👩‍❤️‍👩".truncate_bytes(13, omission: nil)
   end
 
+  def test_truncates_bytes_preserves_encoding
+    original = String.new("a" * 30, encoding: Encoding::UTF_8)
+
+    assert_equal Encoding::UTF_8, original.truncate_bytes(15).encoding
+    assert_equal Encoding::UTF_8, original.truncate_bytes(15, omission: nil).encoding
+    assert_equal Encoding::UTF_8, original.truncate_bytes(15, omission: " ").encoding
+    assert_equal Encoding::UTF_8, original.truncate_bytes(15, omission: "🖖").encoding
+  end
+
   def test_truncate_words
     assert_equal "Hello Big World!", "Hello Big World!".truncate_words(3)
     assert_equal "Hello Big...", "Hello Big World!".truncate_words(2)