Yes. This would actually be working as designed - as long as sufficient fees exist to actually build the tx then all surplus fees can be claimed by the aggregator. In fact they would need to be for the transaction to balance. If the tx itself overpaid and used the surplus fees then others would simply take advantage of this spare capacity and pad the tx out with additional inputs/outputs etc.
Yes absolutely. This would be one way of implementing this in a relatively automated way.
And as long as full transactions took priority then it is trivial to “cancel” the partial transaction if it was decided to use the output before periodic aggregation took place. You simply double spend O1 before aggregation occurs (with an understanding of what that entails privacy wise).
This is a key point to this approach - its effectively free (in aggregate) in terms of on chain storage.