The `HTMLMetadata` class provided two ways to get at the document
metadata:

 1. A `getDocumentMetadata` method which reads the current metadata from
    the document and returns it
 2. A `metadata` field which returns the last-read metadata

The `metadata` field was not used outside the tests and shouldn't be used
because it might return stale metadata (in a web page with client-side
JS that updates `<meta>` and `<link>` tags etc. after the page loads). This field
was also used internally by the various helper methods that gather metadata,
with non-obvious constraints on the order in which the helpers are called.

Remove the field to prevent external mis-use of the class and make the data flow
and dependencies clearer internally.