Merge #14380: fix assert crash when specified change output spend size is unknown

0fb2e69815 CreateTransaction: Assume minimum p2sh-p2wpkh spend size for unknown change (Gregory Sanders)
b06483c96a Remove stale comment in CalculateMaximumSignedInputSize (Gregory Sanders)

Pull request description:

  This is triggered anytime a fundraw type call(psbt or legacy) is used with a change output address that the wallet doesn't know how to sign for.

  This regression was added in 6a34ff5335 since BnB coin selection actually cares about this.

  The fix is to assume the smallest typical spend, a P2SH-P2WPKH, which is calculated using a "prototype" dummy signature flow. Future work could generalize this infrastructure to get estimated sizes of inputs for a variety of types.

  I also removed a comment which I believe is stale and misleading.

Tree-SHA512: c7e2be189e524f81a7aa4454ad9370cefba715e3781f1e462c8bab77e4d27540191419029e3ebda11e3744c0703271e479dcd560d05e4d470048d9633e34da16

test/functional/rpc_psbt.py

@@ -210,6 +210,10 @@ class PSBTTest(BitcoinTestFramework):
             assert tx_in["sequence"] > MAX_BIP125_RBF_SEQUENCE
         assert_equal(decoded_psbt["tx"]["locktime"], 0)
 
+        # Make sure change address wallet does not have P2SH innerscript access to results in success
+        # when attempting BnB coin selection
+        self.nodes[0].walletcreatefundedpsbt([], [{self.nodes[2].getnewaddress():unspent["amount"]+1}], block_height+2, {"changeAddress":self.nodes[1].getnewaddress()}, False)
+
         # Regression test for 14473 (mishandling of already-signed witness transaction):
         psbtx_info = self.nodes[0].walletcreatefundedpsbt([{"txid":unspent["txid"], "vout":unspent["vout"]}], [{self.nodes[2].getnewaddress():unspent["amount"]+1}])
         complete_psbt = self.nodes[0].walletprocesspsbt(psbtx_info["psbt"])