feat(core): implement gater

[iamquang95]

Fix #399

Ports the core/gater component: a DutyGater that rejects duties arriving from peers whose type is invalid or whose epoch is too far in the future.

• crates/core/src/gater.rs — DutyGater::new(client) fetches genesis time + slot config from the beacon node and builds the gater (default budget: 2 future epochs); allows(&duty) -> bool is the predicate, and into_fn() yields the shared DutyGaterFn (Arc<dyn Fn(&Duty) -> bool + Send + Sync>) that the wire components consume.
• Canonical type — pluto_core::gater::DutyGaterFn is now the single source of truth; parsigex, consensus, and dkg use it directly instead of each redefining an identical alias.

[iamquang95]

@claude review this PR

[claude]

Claude finished @iamquang95's task in 2m 20s —— View job

---

Review — feat(core): implement gater

• Read the new gater.rs and changed files
• Compare against Charon Go core/gater.go
• Analyze correctness, style, and tests
• Post review feedback

This is a faithful, high-quality port of Charon's core/gater.go (NewDutyGater). Logic matches the Go original (type validity + future-epoch budget, no past-duty check), the consumer refactor unifying DutyGaterFn is consistent across all crates, and test coverage is excellent (async new/with_options paths, epoch boundaries, pre-genesis, into_fn parity, invalid types). No blocking issues. ✅

Correctness — verified against Go

• is_valid() (crates/core/src/types.rs:70) = !Unknown && !DutySentinel(_), matching Go's t > unknown && t < sentinel — including InfoSync as valid. ✓
• The expect() on checked_div(slots_per_epoch) and checked_div(slot_duration_ms) are sound: fetch_slots_config (crates/eth2api/src/extensions.rs:281) rejects zero slots_per_epoch/slot_duration, and with_options re-checks slot_duration_ms != 0. ✓
• current_epoch() is recomputed per allows() call via the injected clock, mirroring Go calling nowFunc() on each invocation. ✓
• Refactor: no dangling references to the removed pluto_parsigex::DutyGater / qbft::DutyGater aliases anywhere in the workspace — all consumers now use pluto_core::gater::DutyGaterFn. ✓

Intentional divergences (both improvements, worth noting)

1. Pre-genesis clamping. Go computes uint64(currentSlot) from a negative duration, which wraps to a huge epoch and effectively allows everything before genesis. The Rust port clamps elapsed time to 0 (gater.rs:132), so a pre-genesis gater rejects future duties instead. This is more correct and documented as unreachable in practice (gater is built after genesis). Good call — just flagging that it is a behavioral difference from the Go source.
2. Millisecond-resolution arithmetic. Go divides at nanosecond precision (Sub(genesisTime) / slotDuration); the port truncates to whole ms via num_milliseconds() before dividing. Since genesis is whole seconds and slots are seconds-long, this is functionally equivalent (both truncate toward the slot floor); the InvalidSlotDuration guard makes the sub-ms case explicit. No action needed.

Minor / optional

• DutyGater can't derive Debug because of Box<dyn Clock>. Not required here, but if any wiring struct later wants #[derive(Debug)] it'll need a manual impl. Low priority.
• behaviour.rs:25 / handler.rs:22 use single-line nested imports pluto_core::{gater::DutyGaterFn, types::{...}}. Assuming cargo +nightly fmt --all --check passes in CI this is fine; just confirm it's been run.
• The gater() test helper constructs DutyGater via struct literal (fields are private but test is in-module), which nicely sidesteps needing a beacon mock for the synchronous boundary tests. Nice.

Overall: clean port, well-tested, ready to merge from a correctness standpoint. The only thing I'd suggest is a one-line note in the doc comment that the pre-genesis behavior intentionally differs from Charon (in case a future reader diffs the two).
• feat/core-gater