fix(stream_eth): seed the block stream from eth_blockNumber and poll heads alongside newHeads

[DylanVerstraete]

Stacked on #1344 (base is fix/archiver-liveness-a; retarget to usc-dev once that merges). Liveness audit finding 1.

Problem

stream_rpc waited for the first newHeads frame with no deadline before any historical work could start, and the archiver awaited that constructor at startup, in backfill and inside its reconnect loop. A subscription that acknowledges but never delivers (seen through proxies/load balancers, or simply an idle chain) wedged the process indefinitely while /status kept returning 200. The audit reproduced this: "suppress notifications on replacement sockets → stalled at height 24 with no further connection attempts".

Fix

• Seed from eth_blockNumber: catch-up begins immediately and never depends on a future header. Each subscribe/head attempt is bounded by rpc_call_timeout (default 30 s; alloy transports have no default timeout).
• Merged head sources: subscription head numbers are merged with a periodic eth_blockNumber poll (head_poll_interval, default 12 s). Observations that don't advance the frontier contribute nothing, so the two sources overlap freely. The merged stream still ends when the subscription ends, so a dead socket surfaces exactly as before and triggers reconnection.
• expand_heads / merge_heads extracted as pure functions with unit tests (the first tests in stream_eth).
• Archiver: --head-poll-interval-secs / HEAD_POLL_INTERVAL_SECS and --rpc-timeout-secs / RPC_TIMEOUT_SECS, plumbed into all three stream constructions. The attestor uses the same StreamRoots and picks up the defaults unchanged.

Verification

• cargo test -p stream_eth -p archiver: 4 new + 20 passing; clippy -D warnings on stream_eth/archiver/attestor; fmt clean.
• Reproduced the audit scenario with a frame-dropping WS proxy in front of reth --dev (1 s blocks, poll 3 s):

Scenario	Result
Start with eth_subscription frames suppressed	API up at t+12 s, archived 14 (head 16); t+18 s archived 21 → progress while silent
Restore notifications	t+22 s archived 27 = head
Suppress again mid-run	t+30 s archived 33 (head 34) → progress while silent
Watchdog reconnects during the run	0
SIGTERM	graceful stop

Previously the first scenario left the process alive with no API listener and no progress (audit table, last row).

[cursor]

PR Summary

Medium Risk
Changes core ETH head discovery and stream startup on the archiving hot path; behavior is safer when RPC/WS is flaky but mis-tuned timeouts or poll intervals could affect catch-up or reconnect timing.

Overview
Fixes archiver/stream liveness when newHeads acknowledges but stops delivering (or the chain is idle): the ETH root stream no longer blocks on the first subscription frame with no deadline.

stream_eth (stream_rpc) now seeds catch-up from eth_blockNumber right after subscribe, with each subscribe/head attempt bounded by rpc_call_timeout (default 30s). Tip tracking merges newHeads with periodic eth_blockNumber polls (head_poll_interval, default 12s) via new merge_heads / expand_heads helpers; the merged stream still ends when the subscription ends so dead sockets trigger reconnect as before. Poll ticks use Delay missed-tick behavior to avoid RPC floods after long catch-up. Four unit tests cover the new helpers.

Archiver adds --head-poll-interval-secs / --rpc-timeout-secs (and env vars), wired into gap backfill, initial stream, and WS reconnect paths. README documents the flags. Attestor keeps library defaults unchanged.

^{Reviewed by Cursor Bugbot for commit e5ce953. Bugbot is set up for automated code reviews on this repo. Configure here.}

[cursor]

Cursor Bugbot has reviewed your changes using high effort and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 12370dd. Configure here.}