feat: add chunk rpc, upgrade and evm rpc traffic attribution

[jacderida]

Summary

V2-834 Part D.

Part D of the traffic-accounting ground truth. Three new process-global
counter tables, same relaxed-atomic style as the V2-623 replication
table, emitted on the existing 300s summary loop.

• storage::traffic — the largest attribution gap by volume: chunk RPC
  bytes were indistinguishable from DHT messaging in wire_tx_bytes.
  Inbound bytes are counted at ChunkMessage::decode by request kind
  (get / put / quote variants / other / decode_error); outbound bytes at
  send-success in answer_one_request, keyed by kind × outcome (get:
  success/not_found/error; put: success/already_exists/payment_required/
  error) via a traffic_key carried on HandledChunkRequest; failed
  sends itemised separately. chunk rpc traffic summary (cumulative)
  in two groups, target ant_node::storage::traffic.
• upgrade::traffic — archive / binary / signature / manifest response
  body sizes across all three reqwest clients; the releases poll now
  reads bytes() then serde_json::from_slice so the body length is
  observable. upgrade traffic summary (cumulative).
• payment::traffic — EVM RPC call counts and outcomes at the two
  on-chain read sites. Bodies are serialised inside alloy behind a
  fresh per-call evmlib provider, so bytes are not observable from
  ant-node; EVM RPC is HTTPS/TCP and outside the UDP reconciliation
  invariant. evm rpc summary (cumulative).

Linear issue

Closes V2-834

Risk tier

• T0 — docs / tooling / CI / pure UX-output. Repo CI only.
• T1 — client-only, no network-facing behavior change. CI + prod compat smoke.
• T2 — node/client logic with behavioral surface, no protocol/format/economics change. Dev testnet + ADR.
• T3 — protocol / storage format / payments / routing. T2 evidence + adversarial testing.

Counters and INFO lines on the existing replication summary loop. The releases poll now reads the body with bytes() then serde_json::from_slice instead of .json() — same parse, body length made observable.

Compatibility

• Wire: none
• Storage: none
• API: none (new modules are pub(crate); HandledChunkRequest gains a pub(crate) field)

Semver impact

• breaking
• feature
• fix

Test evidence

• cargo clippy --all-features -- -D clippy::panic -D clippy::unwrap_used -D clippy::expect_used (CLAUDE.md): clean
• cargo clippy --all-targets -- -D warnings: clean apart from the pre-existing file_store.rs:2005 redundant_pub_crate lint on main
• cargo fmt --check: clean
• New storage::traffic unit tests (2) pass
• Emitted lines counted against the 32-field cap (largest: 28)

New dependency

none

ADR

n/a

Mitigation / rollback

Revert the PR. Purely additive counters and INFO summary lines at the existing 300s cadence; no config, no wire, no storage change — nothing to migrate.

🤖 Generated with Claude Code

[dirvine]

Approving. Genuinely observability-only: chunk-RPC attribution (rx at decode by kind, tx at send-success by kind×outcome, failed sends itemised), upgrade download body sizes, and EVM RPC call counts. The response.json() → response.bytes() + from_slice refactor is behaviour-equivalent with size checks preserved. Non-blocking note: this sits on the chunk-serving hot path (2 relaxed atomic adds per inbound + 2 per outbound on shared global arrays) — consistent with the existing replication table and dwarfed by chunk decode + ML-DSA verify, but worth revisiting with sharded/thread-local counters if it ever shows up at peak. Note: Security Audit CI failure is pre-existing baseline, not introduced here.