#render_html way slower than #render_doc.to_html

[ojab]

Not sure if it's commonmarker or underlying lib issue, but still.
Highly scientific benchmark

Benchmark.ips do |x|
  10.times do |n|
    x.report("commonmarker_render_html #{n}") do
      CommonMarker.render_html(text, :DEFAULT, [:autolink])
    end

    x.report("commonmarker_two_step #{n}") do
      parsed_doc = CommonMarker.render_doc(text, :DEFAULT, [:autolink])
      parsed_doc.to_html(:DEFAULT, [:autolink])
    end
  end

  x.compare!
end

where text is README.md from this repo shows

Comparison:
commonmarker_two_step 7:     6766.2 i/s
commonmarker_two_step 2:     6753.3 i/s - same-ish: difference falls within error
commonmarker_two_step 8:     6748.9 i/s - same-ish: difference falls within error
commonmarker_two_step 5:     6725.9 i/s - same-ish: difference falls within error
commonmarker_two_step 6:     6720.3 i/s - same-ish: difference falls within error
commonmarker_two_step 1:     6720.3 i/s - same-ish: difference falls within error
commonmarker_two_step 4:     6705.3 i/s - same-ish: difference falls within error
commonmarker_two_step 0:     6703.2 i/s - same-ish: difference falls within error
commonmarker_two_step 9:     6664.8 i/s - same-ish: difference falls within error
commonmarker_two_step 3:     6655.7 i/s - same-ish: difference falls within error
commonmarker_render_html 8:     2542.5 i/s - 2.66x  (± 0.00) slower
commonmarker_render_html 7:     2519.9 i/s - 2.69x  (± 0.00) slower
commonmarker_render_html 3:     2498.4 i/s - 2.71x  (± 0.00) slower
commonmarker_render_html 0:     2492.0 i/s - 2.72x  (± 0.00) slower
commonmarker_render_html 9:     2453.8 i/s - 2.76x  (± 0.00) slower
commonmarker_render_html 2:     2429.4 i/s - 2.79x  (± 0.00) slower
commonmarker_render_html 4:     2412.7 i/s - 2.80x  (± 0.00) slower
commonmarker_render_html 6:     2391.4 i/s - 2.83x  (± 0.00) slower
commonmarker_render_html 5:     2371.8 i/s - 2.85x  (± 0.00) slower
commonmarker_render_html 1:     2363.0 i/s - 2.86x  (± 0.00) slower

so we can easily speedup #render_html 2.5x times by using some more ruby, but I have a suspicion that it should be faster as is.