[bug] Document Fragment #clone does not behave like #dup with respect to the owning document

[flavorjones]

Please describe the bug

While working on upgrading Action Text to HTML5, some tests were failing that I traced back to this repro:

class Minitest::Spec
  describe "HTML4" do
    let(:document) { Nokogiri::HTML4::Document.parse("<div>doc</div>") }
    let(:fragment) { document.fragment("<div>frag</div>") }

    it "#dup makes a copy of the fragment with a new document" do
      dup = fragment.dup
      refute_nil(dup.document)
      refute_same(dup.document, fragment.document)
    end

    it "#clone makes a copy of the fragment with a new document" do
      clone = fragment.clone
      refute_nil(clone.document)
      refute_same(clone.document, fragment.document) ######## this fails ########
    end
  end

  describe "HTML5" do
    let(:document) { Nokogiri::HTML5::Document.parse("<div>doc</div>") }
    let(:fragment) { document.fragment("<div>frag</div>") }

    it "#dup makes a copy of the fragment with a new document" do
      dup = fragment.dup
      refute_nil(dup.document)
      refute_same(dup.document, fragment.document)
    end

    it "#clone makes a copy of the fragment with a new document" do
      clone = fragment.clone
      refute_nil(clone.document) ######## this fails ########
      refute_same(clone.document, fragment.document)
    end
  end
end

fails with

  1) Failure:
HTML4#test_0002_#clone makes a copy of the fragment with a new document [../rails/nokogiri-clone-bug.rb:54]:
Expected #<Nokogiri::HTML4::Document:0x67c name="document" children=[#<Nokogiri::XML::DTD:0x618 name="html">, #<Nokogiri::XML::Element:0x668 name="html" children=[#<Nokogiri::XML::Element:0x654 name="body" children=[#<Nokogiri::XML::Element:0x640 name="div" children=[#<Nokogiri::XML::Text:0x62c "doc">]>]>]>]> (oid=1660) to not be the same as #<Nokogiri::HTML4::Document:0x67c name="document" children=[#<Nokogiri::XML::DTD:0x618 name="html">, #<Nokogiri::XML::Element:0x668 name="html" children=[#<Nokogiri::XML::Element:0x654 name="body" children=[#<Nokogiri::XML::Element:0x640 name="div" children=[#<Nokogiri::XML::Text:0x62c "doc">]>]>]>]> (oid=1660).

  2) Failure:
HTML5#test_0002_#clone makes a copy of the fragment with a new document [../rails/nokogiri-clone-bug.rb:70]:
Expected nil to not be nil.

Specifically:

• HTML4::DocumentFragment#clone re-uses the original's Document
• HTML5::DocumentFragment#clone does not set a document at all!

Expected behavior

The desired behavior matches #dup for node and node-like things: to make a copy of the parent document to avoid accumulating nodes that cannot be freed until the parent document is freed (see #1063 and #1834).

Please also note, while we're in here, that there are expected differences between #dup and #clone that we're also not honoring, see #316 for a description of those problems (essentially, the singleton class). If we tackle this issue, we should tackle that one as well.