Suboptimal AggregationState in DoubleSumAggregation and LongSumAggregation

[weiatwork]

When the aggregated column is of type bigint/double or the like, the partial aggregation ends up creating a wrapped row type for it. This is unnecessary, inefficient and can cause a big increase for the intermediate data. I believe this is due to #464, where the optimization for window functions sacrificed aggregation performance in general. @kokosing @sopel39

explain (type logical)
select ss_item_sk, sum(ss_customer_sk)
from store_sales
group by ss_item_sk;

 Output[columnNames = [ss_item_sk, _col1]]
 │   Layout: [ss_item_sk:bigint, sum:bigint]
 │   Estimates: {rows: 2000 (35.16kB), cpu: 0, memory: 0B, network: 0B}
 │   _col1 := sum
 └─ Aggregate[type = FINAL, keys = [ss_item_sk]]
    │   Layout: [ss_item_sk:bigint, sum:bigint]
    │   Estimates: {rows: 2000 (35.16kB), cpu: 8.39M, memory: 35.16kB, network: 0B}
    │   sum := sum("sum_0")
    └─ LocalExchange[partitioning = HASH, hashColumn = [$hashvalue], arguments = ["ss_item_sk"]]
       │   Layout: [ss_item_sk:bigint, sum_0:row(bigint, bigint), $hashvalue:bigint]
       │   Estimates: {rows: 120527 (8.39MB), cpu: 8.39M, memory: 0B, network: 0B}
       └─ RemoteExchange[partitionCount = 4, type = REPARTITION, hashColumn = [$hashvalue_1]]
          │   Layout: [ss_item_sk:bigint, sum_0:row(bigint, bigint), $hashvalue_1:bigint]
          │   Estimates: {rows: 120527 (8.39MB), cpu: 8.39M, memory: 0B, network: 8.39MB}
          └─ Project[]
             │   Layout: [ss_item_sk:bigint, sum_0:row(bigint, bigint), $hashvalue_2:bigint]
             │   Estimates: {rows: 120527 (8.39MB), cpu: 8.39M, memory: 0B, network: 0B}
             │   $hashvalue_2 := combine_hash(bigint '0', COALESCE("$operator$hash_code"("ss_item_sk"), 0))
             └─ Aggregate[type = PARTIAL, keys = [ss_item_sk]]
                │   Layout: [ss_item_sk:bigint, sum_0:row(bigint, bigint)]
                │   Estimates: {rows: 120527 (7.36MB), cpu: ?, memory: ?, network: ?}
                │   sum_0 := sum("ss_customer_sk")
                └─ TableScan[table = tpcds:store_sales:sf0.01]
                       Layout: [ss_item_sk:bigint, ss_customer_sk:bigint]
                       Estimates: {rows: 120527 (2.03MB), cpu: 2.03M, memory: 0B, network: 0B}
                       ss_customer_sk := tpcds:ss_customer_sk
                       ss_item_sk := tpcds:ss_item_sk