refactor(l1): simplify sync functionality

[0xKitsune]

This PR simplifies finalized L1 synchronization by keeping one next-block cursor across RPC reconnects. It removes queue and tracker cursor reconstruction and keeps the subscription loop directly in run, so partial backfills resume from the first unfinished block.

[0xKitsune]

cyclops audit fast

[github-actions]

cc @0xKitsune

Cyclops audit event published. View workflow run

Config: config: default, iterations: 1, hours: default

[tempoxyz-bot]

👁️ Cyclops Review

This PR keeps a single in-memory L1 sync cursor across reconnects and advances it after each finalized block is fully applied. The cursor-preservation path looks safer for partial retries, but one startup error path now bypasses retry handling, and there is one body-only defense-in-depth note below.

Body-only finding

🛡️ [DEFENSE-IN-DEPTH] Validate fetched header height before applying queue side effects

crates/l1/src/subscriber.rs:602 requests block block_number, but the returned header's own number() is not checked before receipts are fetched by the returned hash and before later side effects use header.number(). A hostile or misbehaving RPC endpoint can return a real finalized header for a later height when the queue is empty; the queue can admit that later height before the tracker rejects it for checkpoint discontinuity, leaving ingestion stuck on retryable out-of-order errors. This appears pre-existing and bounded to liveness because downstream consumers re-check checkpoint continuity.

Recommended Fix: Fail closed immediately after get_header_by_number(block_number.into()) unless header.number() == block_number. This also makes the progress logging arithmetic robust against headers above the intended to range.

Reviewer Callouts

• ⚡ L1 cursor invariant: The refactor removes the previous queue/tracker high-water recomputation and relies on a local next_block. Consider adding a debug assertion near crates/l1/src/subscriber.rs:714 tying next_block to the queue/tracker tips so future changes cannot silently break the invariant.
• ⚡ Backfill progress arithmetic: remaining = to - block_number at crates/l1/src/subscriber.rs:727 is safe only if returned headers cannot exceed the requested range. If height validation is not added, switch this to to.saturating_sub(block_number).

[tempoxyz-bot]

⚠️ [ISSUE] Startup checkpoint read errors now bypass retry handling

run() now calls self.resolve_start_block()? before entering the reconnect/retry loop. Errors from provider.latest() or tempo_num_hash() are classified as retryable Other errors elsewhere, but here they are returned directly to the caller, which panics the critical task. A transient state-provider/storage read failure at subscriber startup can therefore crash the node instead of being retried as it was before this refactor.

Recommended Fix:
Initialize next_block through the same retry policy used for connection/sync errors, or add a small pre-loop retry around resolve_start_block() that retries provider/storage errors while still treating configuration errors such as an unanchored genesis as fatal.