fix(schema): machine-enforce the P-521/ES512 key pairing in profile.json

[vishkaty]

Problem

profile.json encodes the JWK curve↔algorithm binding as schema constraints — but only for three of the four curves it names. There are if/then rules for P-256 → ES256, P-384 → ES384, and Ed25519 → EdDSA, and none for P-521.

The effect, checked with the ucp-schema validator (same one CI uses):

Key published in keys[]	profile.json validation
crv: P-521, alg: ES512 (correct)	✅ Valid
crv: P-521, alg: ES256 (mismatched)	✅ Valid — not caught
crv: P-256, alg: ES384 (mismatched)	❌ Rejected

So a P-521 signing key can be published with an algorithm that contradicts its curve and pass validation, while the identical mistake on a P-256 key is rejected. This is the same class of gap as #571 (a schema-valid artifact depending on a curve/algorithm pairing the schema never defines), at the key-publication layer.

Change

One allOf rule — P-521 pairs with ES512 — mirroring the existing P-256/P-384 rules exactly, plus P-521/ES512 added to the crv/alg examples and the alg "MUST match" description. +10/−5, no reformatting.

It is conditional well-formedness only: it constrains a P-521 key to ES512 when one is published; it does not require any party to support ES512. So it is orthogonal to whether ES512 stays AP2-scoped (#618) or becomes UCP-wide — either way, a P-521 key that is published should be correctly paired.

Verification (ucp-schema)

• The mismatched P-521 key is now rejected; a correct P-521/ES512 key still validates; an alg-less P-521 key still validates (verifiers derive the algorithm from crv); P-256/P-384/OKP keys are unaffected.
• ucp-schema lint source/ (96 files pass), validate_examples.py (276 passed, 0 failed), validator unit tests (43 passed), cspell clean.

Relation to #618

This complements #618 rather than overlapping it — different file, no conflict. #618 documents the AP2 algorithm set and the alg↔curve binding in prose and pseudocode (including a P-521/ES512 JWK example); this makes that binding machine-enforced at the schema layer. I verified that #618's own example JWK validates against this rule, so the two land cleanly together. Happy to defer to @ShuoRen-TT and the maintainers on how to sequence or combine them.

Relates to #571.