If the sender refuses to sign the final transaction, the receiver just propagates the template transaction which pays the receiver! So it's a pretty weak attack.
The only real attack is that the sender could double-spend the template-transaction before it's propagated, but the cost of doing this isn't free, as at the very least you need to pay the transaction fees of creating a double spend. It's not an amazingly good defence, but it's good enough that it's unlikely to get abused (and an attacker would only learn a single utxo of the receiver) .
Okay, I see what you mean. I better understand the weaknesses you've identified, and I can't really think of a better solution than what you've proposed. I also realized that implementors who aren't capable of integrating signing and UTXO validation wouldn't be the ones trying to implement this feature, so my concerns there are also moot. Carry on ;)