fix(evm-charge): remove redundant externalId from Permit2 witness

[chopmob-cloud]

Closes #287.

@Xeift is right: externalId in the Permit2 PaymentWitness is redundant.

The witness commits to challengeHash = keccak256(challenge.id, challenge.realm), and challenge.id is an HMAC whose slot 3 is the entire request object, JCS-serialized per RFC 8785 (core draft-httpauth-payment-00). Since externalId is a member of request, it is already bound to challenge.id (and therefore to challengeHash and the signature) whenever present. The separate witness field binds nothing new.

Removing it is also a small hardening: the old text carried a PaymentWitness.externalId == "" absent-case rule that duplicated, at the witness layer, the absent-optional-field convention the core spec already resolves deterministically at HMAC slot 3 via JCS. Two separately-specified conventions for the same thing is a needless cross-implementation divergence risk.

No regression to the type="hash" mitigation ("Requiring unique externalId values per challenge") — that operates at the challenge.id layer, not the witness, so distinct externalId still yields distinct challengeHash.

Changes (one file):

• Drop string externalId; from the PaymentWitness struct and from the EIP-712 witness type string.
• Reword the Witness Data section to state binding flows through challengeHash -> challenge.id (which already commits to every request field, incl. externalId).
• Update the type="permit2" bullet accordingly.

externalId intentionally remains as the request field, the receipt echo, and the hash-mode mitigation.

Validation: builds clean through kramdown-rfc + xml2rfc (HTML/TXT); rfclint --no-rng --no-spell exit 0; lint_frontmatter.py clean. DCO signed-off.

[brendanjryan]

even if redundant, changing this interface would be breaking. I'm preferential to not do this in the short term