Normalizer assumes all meta tags have a "content" attribute

The NodeJsonOutput tests are currently failing, e.g. https://www.drupal.org/pift-ci-job/752648. Fix this.

The MetatagNormalizer::normalize() method assumes that all meta tags have a "content" attribute because it does this:

foreach ($tags['#attached']['html_head'] as $tag) {
$normalized['value'][$tag[1]] = $tag[0]['#attributes']['content'];
}
Obviously not all meta tags have a "content" element, e.g. LINK tags actually use the "href" attribute.

When the above code is running the $tag array will look something like this:

Array (
    [0] => Array (
        [#tag] => link
        [#attributes] => Array (
            [rel] => canonical
            [href] => http://replacements.mcdev/myaccount/304548
        )
    )
    [1] => canonical_url
)

I think we're going to need to add some logic to load the appropriate class based upon $tag[1] and use it to identify what the "value" is.

Of course this won't work if the meta tag has multiple "values", e.g. MaskIcon.