fix(forkchoice): release mutex during aggregation proof building to prevent consensus stall

[dimka90]

Summary

• Split AggregateCommitteeSignatures() into three lock/unlock/lock phases so the expensive leanmultisig.Aggregate() FFI call (500ms–11s) runs without holding the fork-choice mutex
• Consensus operations (block processing, attestation handling, time advances) continue during proof building
• Block production (produce.go) retains the synchronous locked path via buildAggregatedAttestationsFromSignedLocked() since proofs are needed before publishing

Context

When gean runs as aggregator, AggregateCommitteeSignatures() held the global fork-choice mutex for the entire duration of proof building. With 6 participants this took 11+ seconds, blocking all consensus operations and causing gean to miss 2–3 slots per aggregation cycle. The effect snowballed — missed slots accumulated more attestations, making the next aggregation even slower.

Confirmed by devnet-3 logs and docker stats: gean used 322% CPU as aggregator while non-aggregator clients used under 2%.

After this fix, gean still syncs ~1 block occasionally when aggregation takes 3+ seconds (the proof time is unchanged — that's the upstream leanmultisig library cost), but it no longer falls behind or misses multiple slots. Without --is-aggregator, gean does not sync at all.

Pattern adopted from zeam's separation of signatures_mutex from the forkchoice lock (forkchoice.zig:308).

---

Test Plan

• go build ./... compiles cleanly
• go vet ./... passes
• go test ./chain/forkchoice/... -count=1 — forkchoice tests pass
• go test -race ./chain/forkchoice/... — no data races
• Local 3-node devnet with --is-aggregator — gean stays at head, no multi-slot sync gaps
• Verify aggregation complete duration unchanged (proof time is not reduced, only unblocked)
• Verify behind_peers=false after initial catch-up
• Verify gean can propose blocks while running as aggregator
  Closes fix(forkchoice): release mutex during aggregation proof building to prevent consensus stall #195