Emit the script field in ExtendedStatsAggregationBuilder (re-applies #2700)

[fupelaqu]

What

ExtendedStatsAggregationBuilder never serialises agg.script, although ExtendedStatsAggregation carries the field and the DSL accepts it. This adds the same emission every sibling metric builder already has, mirroring MaxAggregationBuilder's idiom:

agg.script.foreach { script =>
  builder.rawField("script", handlers.script.ScriptBuilderFn(script))
}

Why it matters

The failure is silent: with both a field and a script set, the emitted JSON contains only "field", so Elasticsearch computes the extended stats over the raw field and returns a plausible, wrong number — no error anywhere. Checked against the current sources: ExtendedStats is the only one of the ten metric aggregation builders with this omission (Avg, Cardinality, Max, Min, Percentiles, Stats, Sum, ValueCount and WeightedAvg all emit script).

History — this fix was already approved once

This is a re-application of #2700 (approved and merged on 2022-08-25), which landed on release/7.10.x and was never forward-ported — so the omission is still present in 7.17.25, 8.18.x and series/9.x. The patch here is #2700 adapted to the current handlers.script.ScriptBuilderFn location and the customAggregations signature.

Test

The existing SearchDslTest case "should generate correct json for extendedstats aggregation" already sets Script("doc['grade'].value").lang("lua") on the aggregation — its golden fixture matched the dropped-script output, i.e. the test was pinning the bug. The fixture is updated to expect the script node (same fixture change as #2700).

sbt "tests/testOnly com.sksamuel.elastic4s.search.SearchDslTest -- -z extendedstats"

green locally on JDK 17 / sbt 2.0.8 — and verified red without the builder change (the fixture update alone fails against the unpatched builder), so the test genuinely asserts the emission.

Backport ask

Would you consider a backport to the 7.17 line (a 7.17.26)? We drive ES 6.x–9.x from one SQL engine (SOFTNETWORK-APP/SoftClient4ES#222) on nl.gn0s1s 8.18.x and 7.17.25, and on 7.x our only current option is to reject scripted extended_stats loudly rather than let it answer wrongly. Happy to open the backport PR against the 7.17 branch if you point me at it.

[Philippus]

Nice catch! If you could make the same change towards the 7.x branch series/7.x and the 8.x branch series/8.x that would be great. If not, I can take care of it.

[fupelaqu]

@Philippus Done :

• Emit the script field in ExtendedStatsAggregationBuilder (re-applies #2700) (series/7.x) #4105
• Emit the script field in ExtendedStatsAggregationBuilder (re-applies #2700) (series/8.x) #4106

[Philippus]

@fupelaqu elastic4s 7.17.26 should be released, please try it out.