As pointed out at
https://groups.google.com/forum/#!topic/golang-nuts/LJozHIXAAJY,
`<<p>html</p>` was parsed as `&lt;&lt;p&gt;html</p>`.
There was no test case for this. Chrome parses it as `&lt<p>html</p>`,
and that seems to be correct. We were missing the
"Reconcume the current input character" step at
http://www.whatwg.org/specs/web-apps/current-work/multipage/tokenization.html#tag-open-state

LGTM=nigeltao
R=golang-codereviews, gobot, nigeltao
CC=golang-codereviews, nigeltao
https://golang.org/cl/96060044