[ES|QL] Replace RoundTo linear search evaluator with manual evaluators

[fang-xing-esql]

This is a subtask of #131341.

The RoundToLinearSearchEvaluator does not outperform the manual evaluators, in some cases(data volume dependent), it’s even slower than the DateTruncDatetimeEvaluator.

Below are detailed profiling results (copied from PR #131341). This PR proposes replacing the linear search evaluators with manual ones. If the number of buckets is less than 11, the corresponding manual evaluator will be used instead. This helps performance regression tests when #131341 is merged.

{
            "operator" : "EvalOperator[evaluator=RoundToLongBinarySearchEvaluator[field=Attribute[channel=1]]]",
            "status" : {
              "process_nanos" : 47546640,
              "pages_processed" : 718,
              "rows_received" : 5131188,
              "rows_emitted" : 5131188
            }
},
{
            "operator" : "EvalOperator[evaluator=RoundToLongLinearSearchEvaluator[field=Attribute[channel=1]]]",
            "status" : {
              "process_nanos" : 36383473,
              "pages_processed" : 718,
              "rows_received" : 5131188,
              "rows_emitted" : 5131188
            }
}
{
            "operator" : "EvalOperator[evaluator=RoundToLong9Evaluator[field=Attribute[channel=1], p0=1419811200000, p1=1420416000000, p2=1421020800000, p3=1421625600000, p4=1422230400000, p5=1422835200000, p6=1423440000000, p7=1424044800000, p8=1424649600000]]",
            "status" : {
              "process_nanos" : 24817116,
              "pages_processed" : 718,
              "rows_received" : 5131188,
              "rows_emitted" : 5131188
            }
 }
{
            "operator" : "EvalOperator[evaluator=DateTruncDatetimeEvaluator[fieldVal=Attribute[channel=1], rounding=Rounding[WEEK_OF_WEEKYEAR in Z][fixed to midnight]]]",
            "status" : {
              "process_nanos" : 33711389,
              "pages_processed" : 761,
              "rows_received" : 5699770,
              "rows_emitted" : 5699770
            }
}

The correctness of these manual evaluators is well covered by the existing RoundToTests.
There may be more opportunities to fine-tune performance by experimenting with a wider variety of bucket counts and data volumes. From the current observations, smaller data volumes appear to be more sensitive to the choice of evaluator.

[fang-xing-esql]

Skip the format check on all files under src/main/generated-src/, I don't see this subfolder src/main/generated-src/generated/.

[nik9000]

I liked that we were running spotless on these - it forces us to write the templates in a way that keeps the style consistent. It's a pain though. Can you try removing this and fixing the templates? I know it's really picky and annoying. But it helps keep the code more readable.

[fang-xing-esql]

Thanks for the tricks!

[elasticsearchmachine]

Pinging @elastic/es-analytical-engine (Team:Analytics)

[elasticsearchmachine]

Pinging @elastic/es-perf (Team:Performance)

[elasticsearchmachine]

Hi @fang-xing-esql, I've created a changelog YAML for you.

[nik9000]

I liked that we were running spotless on these - it forces us to write the templates in a way that keeps the style consistent. It's a pain though. Can you try removing this and fixing the templates? I know it's really picky and annoying. But it helps keep the code more readable.

[nik9000]

Could you add a comment to all of the process methods that say that it's basically a hand unrolled binary search?

[nik9000]

Should this be p4 instead?

[fang-xing-esql]

p3 is the median of these 7 numbers.

[fang-xing-esql]

Thanks for reviewing @nik9000! Comments to those manual binary search evaluators are added.