feat(planner_v2): cost function and multi-stage optimization

[wangrunji0408]

This PR adds a practical cost function so that the optimizer can find the best plan.

But since egg doesn't support cost-based heuristic search, it is prone to state explosion and hard to find the best plan as we want in a reasonable time. To solve this problem, we divide the optimization into multiple stages. Each stage will only apply a part of the rules in limited iterations, then extract the best plan as the input of the next stage. In detail, we have 3 stages for now:

1. aggregation extraction and column pruning.
2. predicate and projection pushdown. (3 iterations)
3. join reordering and turning to hash join.

After that, the planner is able to find the best plan for the existing TPC-H queries.

Q1:

Order: [$7.8 asc, $7.9 asc] (cost=15414)
  Projection: [$7.8, $7.9, sum($7.4), sum($7.5), sum(($7.5 * (1 - $7.6))), sum(((1 - $7.6) * ($7.5 + ($7.7 * $7.5)))), avg($7.4), avg($7.5), avg($7.6), rowcount] (cost=15403)
    Aggregate: [rowcount, avg($7.6), avg($7.5), avg($7.4), sum(((1 - $7.6) * ($7.5 + ($7.7 * $7.5)))), sum(($7.5 * (1 - $7.6))), sum($7.5), sum($7.4)], groupby=[$7.8, $7.9] (cost=15390)
      Filter: (1998-09-21 >= $7.10) (cost=12900)
        Scan: [$7.4, $7.5, $7.6, $7.7, $7.8, $7.9, $7.10] (cost=7000)

Q3:

TopN: limit=10, offset=null, orderby=[sum(($7.5 * (1 - $7.6))) desc, $6.4 asc] (cost=54478.105)
  Projection: [$7.0, sum(($7.5 * (1 - $7.6))), $6.4, $6.7] (cost=54473.105)
    Aggregate: [sum(($7.5 * (1 - $7.6)))], groupby=[$7.0, $6.4, $6.7] (cost=54468.105)
      HashJoin: cross, lkey=[$6.0], rkey=[$7.0] (cost=53584.105)
        HashJoin: cross, lkey=[$5.0], rkey=[$6.1] (cost=22651.05)
          Filter: ($5.6 = 'BUILDING') (cost=2700)
            Scan: [$5.0, $5.6] (cost=2000)
          Filter: (1995-03-15 > $6.4) (cost=7500)
            Scan: [$6.0, $6.1, $6.4, $6.7] (cost=4000)
        Filter: ($7.10 > 1995-03-15) (cost=7500)
          Scan: [$7.0, $7.5, $7.6, $7.10] (cost=4000)

Q5:

Order: [sum(($7.5 * (1 - $7.6))) desc] (cost=150338.08)
  Projection: [$0.1, sum(($7.5 * (1 - $7.6)))] (cost=150335.08)
    Aggregate: [sum(($7.5 * (1 - $7.6)))], groupby=[$0.1] (cost=150332.28)
      HashJoin: cross, lkey=[$7.2, $5.3], rkey=[$3.0, $3.3] (cost=149430.28)
        HashJoin: cross, lkey=[$6.0], rkey=[$7.0] (cost=61900.703)
          HashJoin: cross, lkey=[$5.0], rkey=[$6.1] (cost=28966.25)
            Scan: [$5.0, $5.3] (cost=2000)
            Filter: (($6.4 >= 1994-01-01) and (1995-01-01 > $6.4)) (cost=5620)
              Scan: [$6.0, $6.1, $6.4] (cost=3000)
          Scan: [$7.0, $7.2, $7.5, $7.6] (cost=4000)
        HashJoin: cross, lkey=[$3.3], rkey=[$0.0] (cost=51595.125)
          Scan: [$3.0, $3.3] (cost=2000)
          HashJoin: cross, lkey=[$0.2], rkey=[$1.0] (cost=22660.672)
            Scan: [$0.0, $0.1, $0.2] (cost=3000)
            Filter: ($1.1 = 'AFRICA') (cost=2700)
              Scan: [$1.0, $1.1] (cost=2000)

Q6:

Projection: [sum(($7.6 * $7.5))] (cost=7409.828)
  Aggregate: [sum(($7.6 * $7.5))], groupby=[] (cost=7408.328)
    Filter: ((24 > $7.4) and (($7.6 >= 0.07) and ((1995-01-01 > $7.10) and ((0.09 >= $7.6) and ($7.10 >= 1994-01-01))))) (cost=7210.72)
      Scan: [$7.4, $7.5, $7.6, $7.10] (cost=4000)

Q10:

TopN: limit=20, offset=null, orderby=[sum(($7.5 * (1 - $7.6))) desc] (cost=99839.414)
  Projection: [$5.0, $5.1, sum(($7.5 * (1 - $7.6))), $5.5, $0.1, $5.2, $5.4, $5.7] (cost=99830.414)
    Aggregate: [sum(($7.5 * (1 - $7.6)))], groupby=[$5.0, $5.1, $5.5, $5.4, $0.1, $5.2, $5.7] (cost=99821.016)
      HashJoin: cross, lkey=[$5.3], rkey=[$0.0] (cost=98313.016)
        HashJoin: cross, lkey=[$5.0], rkey=[$6.1] (cost=60378.563)
          Scan: [$5.0, $5.1, $5.2, $5.3, $5.4, $5.5, $5.7] (cost=7000)
          HashJoin: cross, lkey=[$6.0], rkey=[$7.0] (cost=23032.313)
            Filter: (($6.4 >= 1993-10-01) and (1994-01-01 > $6.4)) (cost=5620)
              Scan: [$6.0, $6.1, $6.4] (cost=3000)
            Filter: ($7.8 = 'R') (cost=5100)
              Scan: [$7.0, $7.5, $7.6, $7.8] (cost=4000)
        Scan: [$0.0, $0.1] (cost=2000)