feerate of unsigned transaction can be overestimated

[jp1ac4]

This is taken from #1371 (comment).

This issue is about how we estimate the feerate of an unsigned transaction after creation. It's different from #1132 and #1371, which are about how we set the fee when creating a transaction.

The issue concerns the following function:

liana/src/descriptors/mod.rs

Lines 541 to 557 in cbc46e9

	pub fn unsigned_tx_max_vbytes(&self, tx: &bitcoin::Transaction, use_primary_path: bool) -> u64 {
	let witness_factor: u64 = WITNESS_SCALE_FACTOR.try_into().unwrap();
	let num_inputs: u64 = tx.input.len().try_into().unwrap();
	let max_sat_weight: u64 = self.max_sat_weight(use_primary_path).try_into().unwrap();
	// Add weights together before converting to vbytes to avoid rounding up multiple times.
	let tx_wu = tx
	.weight()
	.to_wu()
	.checked_add(max_sat_weight.checked_mul(num_inputs).unwrap())
	.unwrap();
	tx_wu
	.checked_add(witness_factor.checked_sub(1).unwrap())
	.unwrap()
	.checked_div(witness_factor)
	.unwrap()
	}
	}

It seems to estimate the weight to be 2 less than the true weight of the signed transaction. I think this is because we call tx.weight() where all inputs of tx have empty witnesses, and so the transaction is serialized as non-Segwit:
https://docs.rs/bitcoin/0.31.0/src/bitcoin/blockdata/transaction.rs.html#752-753
https://docs.rs/bitcoin/0.31.0/src/bitcoin/blockdata/transaction.rs.html#968-979

This function is used by the GUI to display the estimated feerate of an unsigned PSBT as well as by the daemon to perform sanity checks on a PSBT after creation. By underestimating the weight, we may overestimate the feerate.

The suggested fix is to use tx.weight() + 2 if tx has >= 1 input and all these inputs have empty witnesses.