fix(opstack): subscribe to V4 blocks gossip topic (Base mainnet post-Isthmus)

[amunt0]

Problem

The opstack consensus server is deaf on Base mainnet: it peers fine but never receives a block and never serves a head at /latest.

Root cause (sourced, no guesswork): OP-stack gossips each new block on a versioned topic, and the sequencer publishes to exactly one version — selected by hardfork (op-node p2p/gossip.go):

if p.cfg.IsIsthmus(timestamp) {
    return p.blocksV4.topic.Publish(ctx, out)   // V4 ONLY — no dual-publish
} else if p.cfg.IsEcotone(timestamp) {
    return p.blocksV3.topic.Publish(ctx, out)
} ...

topic	payload	status on Base mainnet
/optimism/8453/2/blocks (V3)	Ecotone-era	dead since 2025-05-09 — what helios subscribes to
/optimism/8453/3/blocks (V4)	Isthmus (withdrawalsRoot)	where every block has been since Superchain Upgrade 15 activated (2025-05-09 16:00:01 UTC)

Upstream a16z master has the same V3-only subscription — this very likely explains why all of a16z's *.operationsolarstorm.org mainnet feeds are down (HTTP 502) while base-sepolia still works.

Fix

7 lines in opstack/src/server/net/block_handler.rs: subscribe to the V4 topic alongside V3 (kept for chains that haven't activated Isthmus). Same sequencer-signed commitment envelope verification on both topics — SequencerCommitment::new + signature check against the network's unsafe_signer are payload-version-agnostic on the server side.

Verification (2026-06-03)

• Without fix: 19 min on --network base, 80 gossip peers, 0 messages, /latest = null
• Topic scan (subscribed v0–v5): first message arrived on /optimism/8453/3/blocks (59,502 bytes) — empirically confirms V4 is the only live topic
• With fix: block received → sequencer signature verified → /latest serves a live signed Base mainnet head → a helios opstack client at the production rev consumed it end-to-end (tracked Base mainnet via /latest + execution RPC)

Also in this PR

• master was fast-forward synced to upstream a16z/helios@1fbb9099 beforehand (standard fork sync, a16z's own code)
• The fork's 4 pre-existing dep-maintenance commits (libp2p 0.56 ≈ upstream chore: update libp2p to 0.56.0 a16z/helios#786, 2× clippy, jsonrpsee features) are re-applied on top so they land on master properly — they were previously dangling off-branch (the production Cargo pin a4fed54 pointed at them)

After merge

1. inco-monorepo bumps helios-wrapper/Cargo.toml rev pin to the merge SHA (1-line PR) → rebuilds helios-wrapper + helios-consensus images
2. This fix should be PR'd upstream to a16z — it fixes their headline opstack use case; once released upstream, this fork can collapse back toward zero delta

[amunt0]

Superseded by the inco integration branch (created at this PR's head, e6d8281 — identical content). New model: master stays a pristine upstream mirror; inco = master + Inco patches (dep maintenance + V4 topic fix) and is what the Docker image builds from; upstream merge of a16z#801 eventually collapses inco back to a mirror.