Replace conversion registration well-known redirect with a response header

[johnivdel]

The API currently supports legacy tracking pixels by allowing them to redirect to a well known url which the browser intercepts. While this works fine, this mechanism is a confusing and abnormal.

Without looking deeply, it is hard to understand how the browser is actually handling the well-known redirect. For example, is it the browser processing the redirect response, and then triggering a conversion? Or is the browser processing the redirect request that triggers attribution?

A more typical mechanism would be having an HTTP header, that when parsed by the browser, is translated to a conversion event.

This could be implemented as a new structured response header of the form:

Register-Conversion: conversion-data=<string>, agg-conversion-data=<string>,...

In the context of #57, the header name could be Trigger-Attribution.

The motivation for this is related to #74, but explicitly simplifies tracking pixel registration.

cc @maudnals @csharrison

[maudnals]

@johnivdel I agree that a header may be an easier-to-understand mechanism than well-known redirects. Maybe this is a topic to bring to the wider group (meeting)?

EDIT: s/would be easier/may be easier

[apasel422]

@johnivdel Is this still something we're considering?

[johnivdel]

I think this has very clear benefits still. We can file a layering issue against PCM as it would increase a large difference in API shape.

[csharrison]

+1 we should align this as best as possible with https://github.com/WICG/conversion-measurement-api/issues/261

[csharrison]

We spoke about this in the meeting on 2021-12-13 (minutes). If we align this we could do something like

<img src="..." attributiontrigger="https://adtech.example/attribution_trigger?id=1234">

This would invoke a fetch just like the attributionsrc requests, and it enables us to make the same security guarantees as attribution sources.

[apasel422]

This has been implemented via Attribution-Reporting-Register-Trigger.