Stabilize CalcitePPLConditionBuiltinFunctionIT on the analytics-engine route

[ahkcs]

Description

Brings CalcitePPLConditionBuiltinFunctionIT to parity on the analytics-engine route (-Dtests.analytics.parquet_indices=true), where every test-created index is composite/parquet-backed and PPL queries route through the analytics engine (DataFusion). The IT already passes on the v2/Calcite route; this change is test-only.

Root cause 1 — append-only accumulation (fixes 12 of 18 failures)

init() seeds two extra docs (_id 7, _id 8) into state_country_with_null via unconditional raw PUTs. init() runs as @Before, before every test method, and preserveClusterUponCompletion() keeps indices across methods. On the analytics-engine parquet-backed store a PUT with an existing _id appends a new row instead of replacing, so the two docs accumulated one duplicate per method and inflated row counts across the suite (the was counts grow monotonically: 13, 16, 25, 26, 32, 36, 42, 45, 47, 50, 52). The bulk loadIndex(...) calls don't have this problem because they're already isIndexExist-guarded. Fix: capture isIndexExist(...) before loadIndex creates the index, then seed the extra docs only on first creation. End state is identical on the v2/Calcite route.

Root cause 2 — four unsupported behaviors on the route (the remaining 6 failures)

These are genuine analytics-engine behaviors, not test bugs (each verified directly against the route). They're skipped via the assumeNotAnalytics(...) registry (AnalyticsRouteLimitation) introduced in #5551, plus matching excludeTestsMatching entries in integTestRemote so the skip set stays countable in one place. NESTED_FIELDS is reused; three new AnalyticsRouteLimitation constants:

• STRUCT_PARENT_FIELD — querying an object/struct parent field directly (isnull(aws) / isnotnull(aws) on big5.aws) resolves to FIELD_NOT_FOUND. The route flattens objects into dotted leaf columns (aws.cloudwatch.log_group scans fine) but the struct parent is not a queryable column. Distinct from NESTED_FIELDS: object parents survive in the OpenSearch mapping but still can't be referenced as a whole. (testIsNullWithStruct, testIsNotNullWithStruct)
• NESTED_FIELDS (reused) — nested_simple.address is type nested, which the route cannot store, so the test infra strips it at index creation ([analytics-engine] Strip AE-unsupported fields from test data and exclude the ITs doomed by it #5541) and the field resolves to FIELD_NOT_FOUND. (testIsNullWithNested, testIsNotNullWithNested)
• CONCAT_NULL_AS_EMPTY — DataFusion concat treats NULL as an empty string (concat('H', null) = 'H'), whereas v2/Calcite propagates NULL (= null), so the null-name row diverges. (testNullIfWithExpression)
• EARLIEST_LATEST_NOW_CLOCK — earliest('now', utc_timestamp()): the relative-time 'now' and utc_timestamp() resolve to the same instant on the route (now >= now → true), but differ on v2/Calcite (false) — a clock-source divergence. (testEarliestWithEval)

Results

Run via :integ-test:integTestRemote against an analytics :run cluster.

Test class	Analytics route — before	Analytics route — after	v2/Calcite route
CalcitePPLConditionBuiltinFunctionIT	6/24 pass, 18 fail	18/18 run, 0 fail (6 excluded)	24/24 pass

On the analytics route the skipped tests are removed by excludeTestsMatching so they don't run; each also carries an in-test assumeNotAnalytics(...) as a safety net. The v2/Calcite route runs all 24 (the gradle excludes apply only to integTestRemote, and assumeNotAnalytics is a no-op when the analytics flag is off).

Check List

• New functionality includes testing.
• New functionality has been documented.
• Commits are signed per the DCO using --signoff.

By submitting this pull request, I confirm that my contribution is made under the terms of the Apache 2.0 license.
For more information on following Developer Certificate of Origin and signing off your commits, please check here.