InstructionProcessor doesn't work in text mode.

If the tokenizer enters 'text mode' then no other tags are parsed. I don't know the spec well enough to know whether it should or not.

Example:

```html
<!DOCTYPE html>
<html lang="<?l en ?>"> <-- not processed
<head>
    <title><?s Title ?></title> <-- not processed
</head>
<body>
<p><?s hello world ?></p> <-- processed
</body>
</html>
```

possibly related: #211