Stabilize CalciteStreamstatsCommandIT on the analytics-engine route

[ahkcs]

Description

Gate the CalciteStreamstatsCommandIT tests that diverge on the analytics-engine route (parquet-backed composite store, DataFusion backend) so the route runs green, while keeping every test active on the v2/Calcite path. Follows the established capability-gating pattern (#5560): an in-test @RequiresCapability(...) annotation plus a matching integTestRemote excludeTestsMatching entry — both no-ops on the v2 route.

Triaged single-shard and multi-shard (num_shards=3) analytics runs against the v2 baseline. Failures fall into three groups:

DOC_MUTATION (4 tests) — testStreamstatsGlobalWithNull, testStreamstatsGlobalWithNullBucket, testStreamstatsResetWithNull, testStreamstatsResetWithNullBucket seed state via PUT+DELETE. Doc-level DELETE is unsupported on the parquet store, and same-_id PUT is append-only, so the leaked doc inflated the row counts of every sibling test reading the shared index. Gated with the same DOC_MUTATION capability the three existing mutation tests already carry.

CHAINED_STREAMSTATS_BY (4 tests) — testMultipleStreamstats, testMultipleStreamstatsWithWindow, testMultipleStreamstatsWithNull1, testMultipleStreamstatsWithEval. Chaining two streamstats where an upstream stage partitions by a group emits a ROW_NUMBER() sequence column from each stage; the Substrait converter names both physical columns identically ("row_number() ROWS BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW"), so the stacked schema has a duplicate/ambiguous field name (500) or, for chained window streamstats, non-deterministic values. The Calcite logical plan is correct (the __stream_seq__ alias is distinct) — the alias is lost in Substrait conversion. A single streamstats by, or a chain where only the final stage has by, works. Fails single- and multi-shard.

STREAMSTATS_SORT_NOT_HONORED (1 test) — testStreamstatsAndSort. The window is computed over the backend scan order, ignoring a preceding | sort (the OVER clause carries no explicit ORDER BY, since streamstats orders by encounter order by design), so the per-row aggregates diverge from the v2/Calcite path.

Pass rate — single-shard analytics route, CalciteStreamstatsCommandIT

metric	before	after
tests run	47	42
failures	12	0
skipped	3	7

The before-failures count is inflated by the DOC_MUTATION leak above; the four leaking tests plus their downstream row-count victims all clear once gated.

v2 route (:integTest): 47 run, 0 failed, 0 skipped — the gates are no-ops off the analytics route.

Out of scope

Twelve further tests fail only on the multi-shard route (they pass single-shard) due to cross-shard fragment-order non-determinism in the streamstats window gather (the coordinator gather has no shard-ordinal tiebreaker, so a preceding | sort can't recover deterministic order). That is an engine-side gap; these are left unchanged rather than gated, to avoid skipping passing tests on the single-shard route.

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.

[dai-chen]

As discussed, let's think how to maintain real backend capability here later. Or maybe there is better mechanism to make our IT work with various engines with different capabilities. Thanks.