This test previously relied on the assumption that the transactions are padded with random data to create two different conflicting transactions. This is a bit fragile, as it would break if for some reason the transaction were already larger than 100 bytes and pad_tx didn't add random data.
Also I am going to need to make the MiniWallet generate deterministic transactions for assumeutxo testing, as we need to checkpoint a deterministic utxo dump. I plan to achieve this by allowing pad_tx to use non-random data for padding, which is another way to break this test.
So use the target_size parameter to explicitely make the transactions different, even if pad_tx adds non-random data. The tx size is arbitrary. I picked a value slightly larger than a standard P2PK transaction (168 bytes), to reduce the risks of pad_tx being a no-op if for any reason we change the wallet mode in the future to RAW_P2PK.
Depends on D17901