Instead of inserting the entire global context into the iframe's HTML,
where we need to encode it to and decode it from JSON and base64, we
only insert a stripped global context, i.e. without the file tree and
the utils. If these are needed in the iframe, for example when a script
changes the DOM and needs to load an image, we use messages to
communicate with the parent document and retrieve the file in question.
It confuses scripts when it is triggered twice. Instead of passing the
global context, we simply write it into the HTML of the iframe.
We need to encode it in base64 to avoid context confusion (e.g. `</script>`
inside the JSON of the global context).
