Split out voting and banking threads in banking stage

Additionally this allows us to aggressively prune the buffer for voting threads
as with the new vote state only the latest vote from each validator is
necessary.

core/benches/banking_stage.rs

@@ -13,6 +13,7 @@ use {
         leader_slot_banking_stage_metrics::LeaderSlotMetricsTracker,
         qos_service::QosService,
         unprocessed_packet_batches::*,
+        unprocessed_transaction_storage::{ThreadType, UnprocessedTransactionStorage},
     },
     solana_entry::entry::{next_hash, Entry},
     solana_gossip::cluster_info::{ClusterInfo, Node},
@@ -80,8 +81,10 @@ fn bench_consume_buffered(bencher: &mut Bencher) {
         let transactions = vec![tx; 4194304];
         let batches = transactions_to_deserialized_packets(&transactions).unwrap();
         let batches_len = batches.len();
-        let mut transaction_buffer =
-            UnprocessedPacketBatches::from_iter(batches.into_iter(), 2 * batches_len);
+        let mut transaction_buffer = UnprocessedTransactionStorage::new_transaction_storage(
+            UnprocessedPacketBatches::from_iter(batches.into_iter(), 2 * batches_len),
+            ThreadType::Transactions,
+        );
         let (s, _r) = unbounded();
         // This tests the performance of buffering packets.
         // If the packet buffers are copied, performance will be poor.