Allow passing in an array of elements_to_score and add 'pre' as a default

[tuzz]

Allow passing in an array of elements_to_score and add 'pre' as a default

We were experiencing a problem where the h1 text was not being included
in the Readability#content. Here is an example that demonstrates the problem:

<article>
  <header>
    <h1>Title</h1>
  </header>
  <section>
    <p>Paragraph</p>
  </section>
</article>

Previously, the code would add the <p>, <section> and <article> elements
as @candidates because it adds the parent and grand parent of every <p>. It
would not add the <header> element as a candidate.

Then, the best_candidate with the highest score is the <section> element.
The code then tries to add related siblings in #get_article but it wasn't
adding the <header> element because it wasn't in the list of candidates.

We can solve this problem by adding <h1> to the list of elements to score
which will then ensure that <header> parent is included in the
candidates and can be added as a related sibling.

This commit also adds <pre> to the list of default nodes to score
because it is included in arc90's original code here:

https://github.com/masukomi/arc90-readability/blob/master/js/readability.js#L749

I'm not sure why this was omitted. Perhaps remove remove code blocks?
Furthermore, this pull request adds a second commit that attempts to solve a follow-on problem:

The code had two strategies for determining whether to include siblings
in the output after determining the best candidate based on score:

1. It checked if the sibling is a candidate that scored above a threshold
   which is the maximum of 10 and 0.2 of the best_candidate’s score.

2. It checked if the sibling was a paragraph that was longer than 80
   characters with a penalty given for each link within the paragraph.

Neither of these strategies worked well for extracting <h1> titles:

1. Failed because titles score poorly due to not containing many commas
2. Failed because titles are within <h1> or <header> elements

However, titles are usually longer than 80 characters and don’t contain
links so it seems reasonable to modify strategy 2) to allow for other
elements, such as <h1> and <header> to be included as related siblings.

Therefore, this commit introduces a :likely_silings option that defaults
to the same <p> elements as before but can now be set by the developer
to include other elements such as <h1> and <header>. These are not
added by default to remain in sync with Arc90’s original implementation.

[cantino]

Thanks @tuzz!

[nattsw]

@cantino will you be releasing a new version of this on https://rubygems.org/gems/ruby-readability? 🤗

[cantino]

Done, 0.7.1.