dev: check @circuitInfo annotations against the compiled k/rows

🧐 Motivation

Every circuit in the repo carries a hand-maintained @circuitInfo k=…, rows=… annotation, and nothing checks it. A refactor that changes a circuit's cost leaves the annotation stale, silently. Reviewers and audit readers treat those numbers as fact, and a k bump is the number that matters most: it doubles proving cost and moves a circuit toward the block and proving limits.

There is no test for this today because the numbers aren't in the artifact. Confirmed by inspection, not assumption:

bzkir headers start midnight:ir-source and prover keys midnight:prover-key[; neither encodes k. Verifier keys are a constant 2119 bytes across every circuit. k can be recovered from prover-key size (size / 2^k lands in 297.9–344.8 against 2× per k step), but that is a constant fitted to a serialization format we don't own.

📝 The data already exists

@openzeppelin/compact-builder already solves the hard part. CompilerService spawns compact under a PTY via script — because compactc only renders the constraint table to a terminal — and Compiler.js parses it and calls writeCircuitInfoJson, producing .circuit-info.json beside the sources:

{
  "generatedAt": "2026-07-27T13:07:56.652Z",
  "files": {
    "MockFungibleToken.compact": [
      { "name": "_approve", "k": 11, "rows": 1409 },
      { "name": "_transfer", "k": 12, "rows": 2192 }
    ]
  }
}

Eight such files exist across contracts/src/** today. So nothing needs to be produced — it needs to be checked.

Two properties of that file shape the design:

• it is written only on a full compile, since k and rows come from the ZK pass (--skip-zk yields only the Compiling N circuits: header)
• it is gitignored (.gitignore:28), so it is a build product with no committed baseline