Add a specialized bulk scorer for regular conjunctions.

[jpountz]

PR #12382 added a bulk scorer for top-k hits on conjunctions that yielded a significant speedup (annotation
FP). This change proposes a similar change for exhaustive collection of conjunctive queries, e.g. for counting, faceting, etc.

[jpountz]

Wikibigall:

                            TaskQPS baseline      StdDevQPS my_modified_version      StdDev                Pct diff p-value
                         Prefix3       88.53      (5.6%)       87.90      (4.5%)   -0.7% ( -10% -    9%) 0.659
                       CountTerm    14431.99      (2.2%)    14391.72      (3.1%)   -0.3% (  -5% -    5%) 0.740
                          IntNRQ      129.57     (19.6%)      129.45     (18.2%)   -0.1% ( -31% -   46%) 0.988
                          Fuzzy1       97.85      (1.3%)       97.83      (0.9%)   -0.0% (  -2% -    2%) 0.961
                          Fuzzy2       79.13      (1.2%)       79.15      (0.8%)    0.0% (  -1% -    2%) 0.921
                        PKLookup      269.31      (1.8%)      269.44      (1.8%)    0.0% (  -3% -    3%) 0.932
           HighTermDayOfYearSort      424.31      (1.6%)      424.69      (1.8%)    0.1% (  -3% -    3%) 0.868
                  CountOrHighMed       85.48     (13.2%)       85.61     (15.9%)    0.1% ( -25% -   33%) 0.975
                      AndHighLow      940.92      (2.0%)      942.38      (2.8%)    0.2% (  -4% -    5%) 0.842
                        Wildcard       67.91      (3.7%)       68.04      (2.8%)    0.2% (  -6% -    6%) 0.855
                 CountOrHighHigh       55.01     (13.4%)       55.14     (16.4%)    0.2% ( -26% -   34%) 0.960
               HighTermMonthSort     4705.02      (2.1%)     4718.66      (1.9%)    0.3% (  -3% -    4%) 0.643
                       OrHighLow      698.53      (1.9%)      700.66      (2.7%)    0.3% (  -4% -    5%) 0.679
                         Respell       60.51      (1.5%)       60.72      (1.4%)    0.3% (  -2% -    3%) 0.458
                      AndHighMed      234.69      (2.1%)      235.53      (2.6%)    0.4% (  -4% -    5%) 0.635
                      HighPhrase        5.54      (7.3%)        5.58      (4.6%)    0.7% ( -10% -   13%) 0.712
                     CountPhrase        3.24     (11.3%)        3.26     (10.1%)    0.7% ( -18% -   24%) 0.830
                     AndHighHigh       51.61      (2.6%)       52.14      (3.4%)    1.0% (  -4% -    7%) 0.279
                       OrHighMed      142.44      (3.4%)      144.55      (4.1%)    1.5% (  -5% -    9%) 0.218
                      OrHighHigh       47.30      (4.1%)       48.17      (4.8%)    1.8% (  -6% -   11%) 0.193
                       MedPhrase       26.15      (5.2%)       26.89      (4.7%)    2.8% (  -6% -   13%) 0.070
                       LowPhrase       66.86      (4.3%)       68.92      (3.7%)    3.1% (  -4% -   11%) 0.016
                         LowTerm      878.77      (5.8%)      919.06      (5.7%)    4.6% (  -6% -   17%) 0.011
                        HighTerm      345.83      (9.5%)      362.55      (8.9%)    4.8% ( -12% -   25%) 0.098
                         MedTerm      532.38      (7.5%)      559.05      (7.0%)    5.0% (  -8% -   21%) 0.029
                 CountAndHighMed      119.60      (3.9%)      126.26      (2.0%)    5.6% (   0% -   11%) 0.000
                CountAndHighHigh       39.81      (4.2%)       43.06      (1.7%)    8.2% (   2% -   14%) 0.000

[jpountz]

This yielded a good speedup on nightly benchmarks. I pushed an annotation.

[jpountz]

Interestingly, it seems to also help with facets: http://people.apache.org/~mikemccand/lucenebench/AndHighHighDayTaxoFacets.html.