Clean up `DocumentMeta#_getLinks` by removing several unnecessary variables and
narrowing the scope of several others.

In the process, avoid trying to normalize the URIs for `<link>` elements
which are going to be ignored anyway because of their `rel` values.

As a side effect this should fix https://github.com/hypothesis/lms/issues/375
because the `<link>` elements shown in https://github.com/hypothesis/lms/issues/375#issuecomment-444637947
will never get parsed. This doesn't close that issue because invalid
URIs could still occur in the remaining `<link>` tags which are parsed.