ART Index: Support compound key scans

[mach-kernel]

🗣 Description

• The ART index already supports compound keys in order to enforce uniqueness constraints.
• ARTKey has a concatenation mechanism that it uses to generate one value key for multiple column exprs.
• This wires it up for the query side, with
  • ARTIndexCompoundKeyScanState, which is the same as ARTIndexScanState but with vectors for an arbitrary number of values
  • ART::CompoundKeyScan, which generates the concatenated key, locks, does an equality scan
• TryScanIndex column binding logic updated with stricter sanity checks and a new unit test

D explain analyze select * from test where x = '525' and y = '2525';
┌─────────────────────────────────────┐
│┌───────────────────────────────────┐│
││    Query Profiling Information    ││
│└───────────────────────────────────┘│
└─────────────────────────────────────┘
explain analyze select * from test where x = '525' and y = '2525';
┌────────────────────────────────────────────────┐
│┌──────────────────────────────────────────────┐│
││              Total Time: 0.0251s             ││
│└──────────────────────────────────────────────┘│
└────────────────────────────────────────────────┘
┌───────────────────────────┐
│           QUERY           │
└─────────────┬─────────────┘
┌─────────────┴─────────────┐
│      EXPLAIN_ANALYZE      │
│    ────────────────────   │
│           0 rows          │
│          (0.00s)          │
└─────────────┬─────────────┘
┌─────────────┴─────────────┐
│         TABLE_SCAN        │
│    ────────────────────   │
│        Table: test        │
│      Type: Index Scan     │
│                           │
│        Projections:       │
│             x             │
│             y             │
│             z             │
│                           │
│          Filters:         │
│          x='525'          │
│          y='2525'         │
│                           │
│           1 row           │
│          (0.00s)          │
└───────────────────────────┘

For completeness (break before passing back to TableScanInitGlobal

---

Benchmark on 7.5M rows. Both have composite index on those columns, but vanilla DuckDB won't push the scan down like this branch does:

With index:

D select * from __data_tdata_1762187237439 where Data2 = '+11118822328' and Data3 = 'D3';
┌──────────────────────┬──────────────────────┬──────────────────────┬───┬───────┬──────────────────────┬───────┬────────┐
│     DateCreated      │     DateUpdated      │        Field1        │ … │ Data7 │        Data8         │ Data9 │ Data10 │
│     timestamp_ns     │     timestamp_ns     │       varchar        │   │ int64 │       varchar        │ int64 │ int64  │
├──────────────────────┼──────────────────────┼──────────────────────┼───┼───────┼──────────────────────┼───────┼────────┤
│ 2024-07-25 13:51:4…  │ 2024-09-27 13:51:5…  │ F1b61faec5-3aed-49…  │ … │   0   │ D80d395e36-9a00-41…  │  821  │   1    │
├──────────────────────┴──────────────────────┴──────────────────────┴───┴───────┴──────────────────────┴───────┴────────┤
│ 1 rows                                                                                            15 columns (7 shown) │
└────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────┘
Run Time (s): real 0.037 user 0.066688 sys 0.008183

No index:

D select * from __data_tdata_1762187237439 where Data2 = '+11118822328' and Data3 = 'D3';
┌──────────────────────┬──────────────────────┬──────────────────────┬───┬───────┬──────────────────────┬───────┬────────┐
│     DateCreated      │     DateUpdated      │        Field1        │ … │ Data7 │        Data8         │ Data9 │ Data10 │
│     timestamp_ns     │     timestamp_ns     │       varchar        │   │ int64 │       varchar        │ int64 │ int64  │
├──────────────────────┼──────────────────────┼──────────────────────┼───┼───────┼──────────────────────┼───────┼────────┤
│ 2024-07-25 13:51:4…  │ 2024-09-27 13:51:5…  │ F1b61faec5-3aed-49…  │ … │   0   │ D80d395e36-9a00-41…  │  821  │   1    │
├──────────────────────┴──────────────────────┴──────────────────────┴───┴───────┴──────────────────────┴───────┴────────┤
│ 1 rows                                                                                            15 columns (7 shown) │
└────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────┘
Run Time (s): real 0.282 user 3.410015 sys 0.047764

[mach-kernel]

This is an update of #7 with refactored column binding logic + additional tests for the new binding mechanism.