fix: prevent deadlock in attestation signature aggregation

[ch4r10t33r]

Summary

Fixes a potential deadlock in attestation signature aggregation by reducing the scope of the signatures_mutex lock.

Problem

The previous implementation held signatures_mutex across onBlock and updateHead calls, which could cause a deadlock:

• Thread A (this): signatures_mutex → forkChoice.mutex
• Thread B (gossip): forkChoice.mutex → signatures_mutex

Solution

Lock signatures_mutex only for the duration of computeAggregatedSignatures, then immediately unlock before calling onBlock/updateHead.

Changes

• Moved signatures_mutex.unlock() to immediately after computeAggregatedSignatures
• Removed defer to make lock scope explicit
• Added detailed comment explaining the deadlock scenario

Test Plan

• Build succeeds
• Tested under concurrent attestation load

[anshalshukla]

It's okay now that we have merged it but I find scoping it better design

{
        self.forkChoice.signatures_mutex.lock();
        defer self.forkChoice.signatures_mutex.unlock();
        const building_timer = zeam_metrics.lean_pq_sig_attestation_signatures_building_time_seconds.start();
        try aggregation.computeAggregatedSignatures(
            attestations,
            &self.forkChoice.aggregated_payloads,
        );
        _ = building_timer.observe();
}

I think the defer would be called as soon as the code section ends