velox: HashAgg w/o spill doesn't outperform vanilla Spark

Description

In a customer workload profiling, we found hashAgg can’t outperform vanilla Spark, even in the case that no spill happens. In this query, the groupby keys are the primary keys of the table so aggregation doesn’t produce reduction at all. Here is the detailed plan stats:

    -- Aggregation[FINAL [n0_0, n0_1] n1_2 := sum("n0_2"), n1_3 := sum("n0_3"), n1_4 := sum("n0_4"), n1_5 := sum("n0_5"), n1_6 := sum("n0_6"), n1_7 := sum("n0_7"), n1_8 := sum("n0_8"), n1_9 := sum("n0_9"), n1_10 := sum("n0_10")] -> n0_0:VARCHAR, n0_1:VARCHAR, n1_2:BIGINT, n1_3:BIGINT, n1_4:DOUBLE, n1_5:BIGINT, n1_6:DOUBLE, n1_7:BIGINT, n1_8:DOUBLE, n1_9:BIGINT, n1_10:DOUBLE
       Output: 28319541 rows (6.01GB, 6914 batches), Cpu time: 37.48s, Blocked wall time: 0ns, Peak memory: 5.05GB, Memory allocations: 207459, Threads: 1
          distinctKey0                 sum: 1, count: 1, min: 1, max: 1
          distinctKey1                 sum: 32250, count: 1, min: 32250, max: 32250
          hashtable.capacity           sum: 33554432, count: 1, min: 33554432, max: 33554432
          hashtable.numDistinct        sum: 28319541, count: 1, min: 28319541, max: 28319541
          hashtable.numRehashes        sum: 14, count: 1, min: 14, max: 14
          hashtable.numTombstones      sum: 0, count: 1, min: 0, max: 0
          runningAddInputWallNanos     sum: 34.32s, count: 1, min: 34.32s, max: 34.32s
          runningFinishWallNanos       sum: 1.62us, count: 1, min: 1.62us, max: 1.62us
          runningGetOutputWallNanos    sum: 3.15s, count: 1, min: 3.15s, max: 3.15s

ping @FelixYBW @mbasmanova.