[BIP-174] Clarify that partial_sig should be a valid.

In discussing https://github.com/rust-bitcoin/rust-bitcoin/pull/669?notification_referrer_id=NT_kwDOAA2G-7EyNDgwMTU5NzIxOjg4NjUyMw#event-5861680619 it's clear that the PSBT spec is ambiguous with respect to what values should be parseable for PSBTs, this patch clears up the ambiguity.

---

Personally, I feel that the better path is to allow NULLDUMMY here, as there could be some situations whereby satisfaction is NP-hard without some indicator of use a signature or not:

e.g., a contrived example (not saying you should do this) is

<0> toaltstack
<pk0> Checksig OP_IF FROMALTSTACK <V0> OP_ADD TOALTSTACK OP_ENDIF
<pk1> Checksig OP_IF FROMALTSTACK <V1> OP_ADD TOALTSTACK OP_ENDIF
...
<pkN> Checksig OP_IF FROMALTSTACK <VN> OP_ADD OP_ENDIF
<T> OP_EQUAL

where to satisfy requires solving a subset-sum, so just collecting all valid signatures does not allow a polynomial time finalizer to solve by inference in many cases even if all signers add their signature.

Of course, this isn't the best example, but it is an example nonetheless.