fix(aggregation): resolve signers from the head state's validator registry - #420

Step 2 of #418: remove the failure mode that step 1 made visible.

Stacked on #419 — review that first; this PR's diff is against it.

The problem

aggregateFromSnapshot looked up a stored state per attestation target and dropped the group when that state was absent:

targetState := snap.targetStates[attData.Target.Root]
if targetState == nil {
    return
}

On devnet-5 (2026-08-31, 14:35–15:04 UTC) that was every group, every slot. gean_7 produced nothing in 355 of 356 slots while healthy (Behind: 0–3) and holding 753 signatures, each session finishing in 14–90 µs. It held 262 states against a head of slot 399, and the votes it was aggregating carried target slot 342.

Why the lookup was never needed

Every use of that state was state.Validators[...] — resolving attestation pubkeys, nothing else.

The validator registry is written once, in generate_genesis, and never by the state transition. I checked the pinned spec (lstar/state_transition.py): validators= appears only in genesis construction, and gean's internal/statetransition never assigns to .Validators either.

So every state on the chain carries the same registry, and the head state resolves exactly the same signers. SnapshotInputs already requires a head state and returns nil without one, so it is always available — unlike an arbitrary target's state.

What changes

• Signers resolve against snap.headState.Validators.
• The per-target state map and the store read per data root that filled it are gone.
• The missing_target_state skip reason is removed: the condition it named can no longer occur.

Selection, ordering, budget and the resulting aggregates are otherwise unchanged.

How ethlambda does it

crates/blockchain/src/aggregation.rs builds one ProjectedState from the head state and validates candidates against that chain view. It has never needed a per-target stored state; this brings gean to the same footing.

Risk

The change rests on the registry being genesis-fixed. If a future spec adds validator set changes, resolving against the head would be wrong for a target under a different registry — the reasoning is recorded in a comment at the call site so it surfaces during that work.

Testing

• make test — all 25 packages green; make lint, go vet clean; -race clean on aggregation, node, store.
• New test asserts a group whose target has no stored state now reaches signer selection instead of being dropped.
• Two existing fixtures needed correcting, and the reason is worth noting: aggregateTestSnapshot built a snapshot with no head state and votes targeting slot 0 — neither of which SnapshotInputs can produce. They now carry a head state and a target slot matching the vote slot.