FLEDGE - interest_group object data in report_win

[mbowiewilson]

In FLEDGE, it appears that generate_bid does have access to the interest_group object, but report_win does not. This poses a major problem for buyers hoping to leverage machine learning to generate bids because any signal unavailable to report_win cannot be used as a feature in a machine learning pipeline, even if it is available to generate_bid. Without feedback on how the ads perform for different states of the interest_group object we don't have an effective way of leveraging the interest group information for bidding.

To make this issue more concrete, we were hoping to train a machine learning model to help us bid on ad opportunities using the data sent to us from report_win. Model inference in FLEDGE is ugly, but possible: we could "partially query" the model during the interest group request and "partially query" the model during the contextual request, and then stitch these partial query results together in the generate_bid function. However, for this type of flow to work we would need to be able to log information from the interest_group object so this model can have interest group data over which to train.

[mbowiewilson]

After this PR #303 we have a long term plan for resolving this issue. Before the Private Aggregation API comes into service though this issue is still relevant.

[michaelkleber]

And in the short term (like for initial testing right now!), forDebuggingOnly.reportAdAuctionWin() and forDebuggingOnly.reportAdAuctionLoss() (described in the First OT Details doc) get things off the ground.