fix(keystore): fixed-width ECDSA r/s and X/Y encoding (leading-zero bug) + CI that actually runs

[amirradjou]

Summary

Fixes the ~1.5% random signature / public-key verification failures that made the test suite flaky and silently dropped ~1-2% of log entries, then makes CI actually run and gate.

big.Int.Bytes() strips leading zero bytes, so about 1 in 128 P-256 signatures (r || s) and 1 in 128 public keys (X || Y) were encoded 63 bytes long. VerifyMessage split those at len/2 and ReconstructPublicKeyFromHex rejected them (invalid public key length: 63), so freshly created identities could not verify their own signatures and every Log read path (Get, Values, Traverse, JoinEntry) dropped the affected entries with Skipping entry with invalid signature.

What changed

• keystore (keystore/key-store.go): encode r/s and X/Y with big.Int.FillBytes into fixed 32-byte halves (CoordinateSize, SignatureSize, PublicKeySize); new EncodePublicKey is the single public-key encoder; VerifyMessage and ReconstructPublicKeyFromHex now require the exact length instead of "at least 64, split in the middle".
• identities / providers: CreateIdentity uses keystore.EncodePublicKey; VerifyIdentity and Identities.Verify use keystore.ReconstructPublicKeyFromHex instead of their own midpoint-split parsers.
• Regression tests (all fail against the old encoding, verified by temporarily re-introducing it):
  • keystore: 3000 fresh keys sign -> verify with fixed-width assertions; deterministic vectors with a known leading-zero X, Y, r and s (fixed-width verifies, the legacy 63-byte form is rejected); zero-padding unit test; malformed-length rejection.
  • identities: 3000 identities Create -> VerifyIdentity -> Sign -> Verify; malformed public-key rejection.
  • identities/providers: known short-X key through CreateIdentity/VerifyIdentity; 3000-identity loop.
  • oplog: 1000 fresh identities NewEntry -> VerifyEntrySignature (+ CBOR round trip); 300 Log.Append -> Get/Values read-backs.
• chore(lint): clear the four pre-existing staticcheck findings (two unused mu fields, an unused test helper, a no-op type assertion) so staticcheck can gate.
• ci: runs-on: ${{ matrix.os }} (ubuntu/macos/windows at the go.mod floor 1.23, plus ubuntu on stable), go mod download + build + go test -race -count=1 -coverprofile + a -count=20 loop over the crypto packages, upload-artifact@v4; a separate lint job with go mod verify, go mod tidy -diff, gofmt, go vet, staticcheck and govulncheck via go run ...@latest; permissions: contents: read; concurrency cancel; .github/dependabot.yml (gomod grouped + github-actions).

Verification (local, Go 1.26.8, linux/amd64)

Before (upstream/main @ 1064195):

$ go test -count=50 ./keystore/ ./identities/... ./oplog/
ok    orbitdb/go-orbitdb/keystore
FAIL  orbitdb/go-orbitdb/identities             (TestVerifyIdentity x1, TestSignAndVerify x1)
ok    orbitdb/go-orbitdb/identities/identitytypes
FAIL  orbitdb/go-orbitdb/identities/providers   (TestVerifyIdentity x1, TestCreateIdentity x2)
FAIL  orbitdb/go-orbitdb/oplog                  (TestLog_Values x3, TestLog_AppendAndGet x3, TestLog_Join, TestLog_Traverse, TestVerifyEntrySignature)
=> 3 of 5 packages FAIL, 14 test-case failures

$ go test -race -count=1 ./...
7 packages ok, FAIL syncutils (TestSyncSendAndReceiveHead, orbit-sync_test.go:422 SyncedCh ordering), 0 DATA RACE warnings

After (this branch):

$ go test -count=50 ./keystore/ ./identities/... ./oplog/
ok    orbitdb/go-orbitdb/keystore                  15.812s
ok    orbitdb/go-orbitdb/identities                53.084s
ok    orbitdb/go-orbitdb/identities/identitytypes   0.007s
ok    orbitdb/go-orbitdb/identities/providers      30.768s
ok    orbitdb/go-orbitdb/oplog                     17.799s
=> 5 of 5 ok, 0 failures

$ go test -count=1 ./...            => all 8 packages ok
$ go test -race -count=1 ./...      => 7 packages ok, FAIL syncutils (same pre-existing TestSyncSendAndReceiveHead ordering flake; fails 4/5 runs on untouched upstream/main too), 0 DATA RACE warnings
$ go build ./... && go vet ./... && gofmt -l . && go mod tidy -diff   => clean
$ go run honnef.co/go/tools/cmd/staticcheck@latest ./...             => clean (was 4 findings)
$ go run golang.org/x/vuln/cmd/govulncheck@latest ./...              => 2 reachable (see below)

The TestSignVerifyManyFreshKeys log line from one run: 3000 keys: 30 had a leading-zero coordinate, 25 signatures had a leading-zero r or s (all verified).

Known red in the new CI until follow-ups land

• govulncheck (lint job): GO-2026-5073 and GO-2026-5684 in go-ipld-prime v0.21.0, reachable from oplog.Decode. The fix is go get github.com/ipld/go-ipld-prime@v0.23.0 && go mod tidy, but v0.23.0 declares go 1.25.7, so it raises the module's Go floor from 1.23 (and the CI matrix version). Left out of this PR as a separate decision.
• syncutils TestSyncSendAndReceiveHead under -race: pre-existing ordering flake (peer-join event arrives on SyncedCh before the Add payload), unrelated to this change.

Note: the P-256 FillBytes fix is an explicit stopgap; the planned move to secp256k1 via go-libp2p core/crypto (DER signatures, compressed keys) replaces this code entirely.

---

Opened as a draft from the amirradjou fork; identical to amirradjou#1 (merged there). Happy to split the CI commit into its own PR if preferred.