perf(engine): skip dispatch pipeline when all proof targets already fetched

Add early return in on_hashed_state_update when not_fetched_state_update
is empty, avoiding unnecessary MultiAddedRemovedKeys clone, Arc
allocation, dispatch_with_chunking call, and proof target construction
for state updates where all accounts and storage slots were already
prefetched.

For small blocks (<50 txs) where prewarming runs ahead of execution,
many state updates will have all targets already fetched. Each
already-fetched update saves ~2-5µs by skipping:
- AddedRemovedKeys HashSet clone (~200-500ns)
- Arc allocation + later drop (~100-200ns)
- dispatch_with_chunking → get_proof_targets → dispatch → channel
  round trip (~500ns-2µs)
- ProofSequencer bookkeeping (~100ns)
- MultiProofTargets::default() + extend (~50-100ns)

Estimated savings for 30-tx block with ~20 prefetched updates: ~40-100µs

Amp-Thread-ID: https://ampcode.com/threads/T-019c5151-b4af-7475-a7d9-8cfa8f1d22aa

crates/engine/tree/src/tree/payload_processor/multiproof.rs

@@ -889,6 +889,12 @@ impl MultiProofTask {
             state_updates += 1;
         }
 
+        // Fast path: if all targets were already fetched, skip the expensive
+        // MultiAddedRemovedKeys clone, dispatch_with_chunking, and proof target construction.
+        if not_fetched_state_update.is_empty() {
+            return state_updates;
+        }
+
         // Clone+Arc MultiAddedRemovedKeys for sharing with the dispatched multiproof tasks
         let multi_added_removed_keys = Arc::new(MultiAddedRemovedKeys {
             account: self.multi_added_removed_keys.account.clone(),