Handle decoding of input in html5ever

[simonwuelker]

These changes are an attempt to allow users of html5ever to respect the encodings specified with <meta charset="..."> tags in a spec-compliant way.

The major change is that the https://html.spec.whatwg.org/#input-stream now lives in the html5ever crates (if the decoding wrapper around the tokenizer is used). As a result, the new API surface exposes a "pull" instead of the existing "push" interface.

The entry point to the new API is a DecodingParser, which wraps either a HTML or an XML parser.
After providing some amount of byte input to a DecodingParser, the user can call DecodingParser::parse, which returns an iterator over ParserActions. A parser action is either a <script> tag that needs to be executed or a new encoding that the document should be re-parsed with. The caller can drive the parser by repeatedly advancing this iterator.

The old API is fully preserved, without breaking changes (that I'm aware of).

This is a draft because the design is not final and this needs a companion servo PR to verify the correctness of these changes. Initial feedback is welcome.

Depends on #591.

[simonwuelker]

The major change is that the https://html.spec.whatwg.org/#input-stream now lives in the html5ever crates (if the decoding wrapper around the tokenizer is used). As a result, the new API surface exposes a "pull" instead of the existing "push" interface.

I am running into architectural issues with this approach, because the prefetch tokenizer needs to run on the same input as the parser, but it shouldn't need to independently decode the incoming bytes. And since the prefetch tokenizer lives in servo, that's going to be difficult when when we decode input in html5ever.
I think it will be easier to keep the decoding in servo.