feat(network): add Zakura block sync

[evan-forbes]

Motivation

Final stacked PR 5/5 of the Zakura networking layer. Builds on PR4. Adds Zakura block sync (download/verify/commit of block bodies over Zakura, coexisting with legacy sync), plus the security hardening, state/sync-flow refactor, and stabilization/throughput work that complete the feature.

Solution

Organized into four reviewable commits:

1. feat(network): add Zakura block sync — block-sync wire + service scaffold, header v2 and body size hints, download reactor + scheduler, verify/commit/legacy coexistence, chain-tip reset handling, body serving/advertising, native-discovery integration, and the block-sync e2e + header-store bootstrap.
2. fix(network): harden Zakura P2P against security audit findings — supervised-pipe abstraction and resolution of audit findings across transport, handshake, legacy gossip, discovery, header-sync, and block-sync (rate/alloc caps, per-IP caps, dial/task-lifetime supervision, malformed-prelude handling).
3. refactor(network): rework Zakura block/header sync state flow — block/header sync state-management and commit-flow rework (adopts upstream's try_handoff_to_non_finalized_write checkpoint→non-finalized handoff).
4. fix(network): stabilize Zakura block sync and tune throughput — final stabilization plus raising block-sync request size to 128 blocks and using the byte budget as the congestion window.

Tests

• zebra-state (185 tests) and zebra-network suites pass locally.
• cargo clippy --workspace --all-targets and cargo fmt --all -- --check clean locally with the CI feature set.
• Zakura regtest e2e (cargo test -p zebrad --test zakura_regtest_e2e -- --ignored) exercises the full sync path.

Specifications & References

• Overlaps with @p0mvn's standalone PRs fix: pipeline block-sync body commits so checkpoint ranges finalize #88 (pipelined block-sync commit) and fix: rate-limit redundant header-sync status instead of disconnecting #89 (rate-limit redundant header-sync), which were extracted from related work — coordination needed at merge time.

AI Disclosure

• AI tools were used: Claude Code (Opus) for rebasing onto ironwood-main, resolving conflicts (notably the zebra-state handoff/pruned-storage merge), squashing 68 commits into these four, and this description.

PR Checklist

• Conventional-commit title
• Solution is tested
• Changelog updated

[evan-forbes]

Superseded by a 3-PR restructure: only commits at clean boundaries (C1/C5/C8) compile + pass clippy standalone, so the 5-way split was collapsed into #106 (P2P stack), #111 (discovery + header sync + block sync), and #112 (block-sync hardening/refactor/stabilization).