feat(commonware): upgrade to 2026.5.0

[erenyegit]

What this does

Migrates the kora workspace from commonware 2026.4.0 to the 2026.5.0 release.

Why it's a draft

kora-e2e tests::consensus::test_empty_blocks (and any e2e test that uses TestHarness::run) overflows the test thread stack post-upgrade. The build is clean and the trusted-devnet smoke run finalizes, but this is a real regression I want to leave visible while we decide on the right fix. See the "Known issue" section below.

Major API migrations

• Storage: merkle::journaled → merkle::full. QMDB VariableConfig and Db gain S: Strategy generic (using Sequential).
• Runtime: tokio::Context no longer implements Clone. with_label and clone replaced with Supervisor::child("label"). Metrics::register now returns Registered<M>.
• Reporter / Blocker / Manager: now sync, return commonware_actor::Feedback instead of impl Future. Manual Clone impls on FinalizedReporter, SeedReporter, and CommonwareRootProvider use context.child("xxx_clone").
• Application trait: genesis removed. verify moved into Application (VerifyingApplication trait removed). propose/verify take impl Ancestry<Self::Block> instead of AncestorStream<A, B> with BlockProvider.
• Config additions: simplex::Config gained floor: Floor<S, D>; marshal::Config gained start: Start<P::Scheme, B::Digest, B>. ActorInitializer::init / init_with_partition now take a start parameter.
• Net: Sender::send returns Vec<PublicKey>; .await dropped on sync API calls.
• Ed25519: internals are vendored. PrivateKey constructed via ReadExt::read (was ed25519_consensus::SigningKey::from(seed)).
• DKG: bls12381::dkg types moved under the feldman_desmedt submodule.

Cross-referenced will's fix/commonware-resolver-upgrade branch on Nunchi-trade/daeji as a migration reference. simplex config/engine, marshal peers/broadcast, and transport-sim are copied wholesale from there.

Validation

• cargo build --workspace clean
• cargo test --workspace --no-run compiles all test targets
• kora-marshal integration tests pass
• just trusted-devnet brings up 4 validators + 1 secondary peer all healthy and finalizes at around 3 blocks/sec on this (I/O-degraded) host

Known issue: e2e stack overflow

tests::consensus::test_empty_blocks and any other test that goes through TestHarness::run overflows the test thread's 8MB stack.

Likely cause: our manual Clone impls on FinalizedReporter, SeedReporter, and CommonwareRootProvider call context.child("xxx_clone") because released 2026.5.0 no longer has Context: Clone. Over the test's event-loop iterations the supervision tree appears to grow until the default stack is exhausted.

daeji avoids this by patching commonware to main (where Context: Clone still exists). Three follow-ups under consideration:

1. Refactor the Clone impls to use Arc<Context> or similar so cloning doesn't grow the supervision tree.
2. Match daeji and apply [patch.crates-io] to point commonware at main.
3. Bump RUST_MIN_STACK for the test binary.

Per request, not marking the failing test #[ignore] so the regression is visible. For context, 18 other kora-e2e tests are already #[ignore]'d as pre-existing flaky-in-parallel.

Follow-ups (separate PRs)

• Resolve the stack overflow (one of the three options above).
• Stability + tx load pass with the new mempool (spammoor-style).