Crypto surface (keygen/encrypt/decrypt/sign) has 0% coverage and 17/17 UNKNOWN hardware #2590

The cryptographic surface has zero test coverage and zero hardware verification.

Measured, from docs/audits/surface_audit_clustered.csv

cluster 13  orchestrate-pacha-secrets    17 features    0.0% covered    17/17 UNKNOWN hardware

The features include keygen, encrypt, decrypt, sign.

Every other zero-coverage cluster is uncovered surface. This one is uncovered
crypto — and unlike contracts-pv (0 UNKNOWN, so at least hardware-verified), here
all 17 are both ungated and unverified. Nothing has confirmed these work on any
machine.

Why crypto is not just another cluster

For most features, "no gate" means a defect ships and someone notices. For key
generation, signing and encryption, the failure modes are silent by construction:

• a keygen producing low-entropy or deterministic keys looks identical to a correct one
• an encrypt/decrypt round-trip that passes proves nothing about whether the
  ciphertext is actually secure
• a sign that verifies against its own verifier proves nothing about interoperability
  or about rejecting a forged signature

These do not announce themselves in a dogfood sweep the way a 404 or a panic does. A
sweep that exercises them and reports success is the weakest kind of evidence, and
the one most likely to be mistaken for the strongest.

What coverage has to mean here

Not "the command runs." At minimum:

• keygen → distinct keys across invocations, and a known-answer test where the
  algorithm admits one
• encrypt/decrypt → round-trip plus a negative: a tampered ciphertext must FAIL,
  not silently return garbage
• sign → verifies a valid signature and rejects a forged one. A verifier that
  accepts everything passes the positive test perfectly.
• at least one host in the matrix must move verified_hardware off UNKNOWN

Each of those negatives is the discrimination check this repo already applies elsewhere:
a gate that is GREEN for both the correct and the tampered input is not measuring what it
claims.

Note on priority

I am not claiming a defect exists here — no one has looked, which is the point. Given
17 features, all UNKNOWN, in a domain where failure is silent, this is the cluster I would
gate first among the nine zeros, ahead of larger ones like http-orchestrate-banco (95
features) where failures are loud.

Refs #2582