[flink][spark] Support composite BTree global index creation

[JingsongLi]

Purpose

Allow Flink and Spark to build composite BTree global indexes for Data Evolution tables through sys.create_global_index with an ordered, comma-separated column list.

This follows the storage and equality-query support merged in #10338 and #10339.

• Pass every key column to the index scanner, writer and indexer.
• Construct typed tuple keys in both engines and preserve key order during sorting.
• Keep independently configured single-column indexes and composite definitions separate.
• Add integration coverage for nullable components, reader/eager equality queries, incremental builds, updates to either component, unchanged rebuilds and precise index removal.
• Document the current full-key equality contract and build lifecycle.

Python support remains deferred. Prefix, range and IN query optimization will follow separately.

Tests

All scoped tests pass without the fast-build profile:

Profile	Runtime	Scope	Result
Flink 1	JDK 8	Composite/single BTree, Bitmap, MultiValue integration tests; topology and procedure unit tests	15 passed
Flink 2	JDK 11	Same Flink scope after rebuilding the common module	15 passed
Spark 3	JDK 8 / Scala 2.12	Composite and existing create-index procedure suites; topology and procedure Java unit tests	9 Scala + 15 Java passed
Spark 4	JDK 17 / Scala 2.13	Same Spark scope after a clean dependency build	9 Scala + 15 Java passed

Checkstyle, Spotless, Maven Enforcer, Apache RAT and git diff --check also pass.

[leaves12138]

Reviewed 08c3521. No blocking issues found.

Checked the ordered field propagation, tuple materialization and sorting in both engines, coexistence with independent single-column definitions, and the incremental build / component-update / drop lifecycle.

Local validation on JDK 8, without fast-build:

• Flink 1: all 15 scoped integration/topology/procedure tests passed.
• Spark 3 / Scala 2.12: all 9 Scala procedure tests and 15 Java topology/procedure tests passed.
• An additional temporary Flink integration test passed with 6,000 rows across two partitions, a three-column key in non-schema order, object reuse enabled, and a 64 KiB sort buffer. It covered partition-filtered construction followed by incremental construction of the remaining partition and equality lookups in both eager and reader modes.

Full GitHub CI still has pending jobs at review time; Flink 2 and Spark 4 were not rerun locally.