nut10: Add witness section and warning

[robwoodgate]

Mints currently treat Proofs with a witness field but a plain (non-NUT-10) secret differently.

Nutshell rejects, CDK ignores the witness.

This PR codifies this behavior with a warning.

[robwoodgate]

@ye0man - This PR still highlights a real problem, but this far down the road, I think the real solution is to tweak Nutshell to ignore witnesses it doesn't need (follow CDK).

With keysets v3 on the horizon, there will be plenty of non-NUT10 shaped secrets that DO require a witness so the comment, as written, will age badly.

[KvngMikey]

With the proposal for Nutshell to adapt CDK's behavior (ignore witnesses it doesn't need), I think the normative text in this PR should flip from "MAY reject" to "SHOULD ignore":

Proof.witness is only expected for proofs containing a NUT-10 Well-known Secret.
A mint SHOULD ignore a witness field on a Proof whose secret cannot be parsed as a Well-known Secret.

Worth keeping/adding alongside this:

• Mints SHOULD still reject malformed JSON secrets that declare a NUT-10 shape or known kind (genuine malformedness, not plainness).
• Mints MUST still reject plain secrets inside SIG_ALL batches, where every input must be a spending condition.
• Optionally: mints MAY log when they drop a witness on a plain secret, so the behavior is observable without being a hard failure.