HTML parser has a lot of overhead for speculative tokenizing. This is only necessary for document.write.

If document.write is disabled for the document, we can assume all tokens are correct, and we can stop sending / storing checkpoints all over the place in the parser code.

I may do some perf/memory testing on some big sites to see if this is a win. If so, this could be a nice carrot to encourage sites to use the feature policy here.