Skip max score computation when sorting by relevance in BlockGroupingCollector

[gaobinlong]

Description

In BlockGroupingCollector.getTopGroups(), we firstly compute the groupMaxScore by iterating on each group docs to get the max value, then compute the totalMaxScore by Math.max(maxScore, groupMaxScore), both can be optimized when the sort is by relevance.

For the groupMaxScore, we can get the maxScore from the top scoreDoc in the merged top docs. For the totalMaxScore, we can get it from the groups[0] directly if sort by relevance.

Performance benchmark:

  TaskQPS baseline      StdDevQPS my_modified_version      StdDev                Pct diff p-value
                    TermBGroup1M       63.31      (9.3%)       64.90      (6.5%)    2.5% ( -12% -   20%) 0.325
                  TermBGroup1M1P      203.86     (12.9%)      211.23     (18.2%)    3.6% ( -24% -   39%) 0.468

[gaobinlong]

Hi @benwtrent , please help to review this PR when you're convenient, thanks!

[benwtrent]

simple enough, I like it