[nitinware]

Adds CouchbaseIO, a Java connector for Couchbase Server. Fixes #18381.

Design discussion: https://lists.apache.org/thread/cvzpfnw717zbnf7z6slfkfbp6zg6604g

Design doc: https://docs.google.com/document/d/1ALlXaiVSa_Dn9jJxM9MsmhEhBLPW_FP8RswN8HG33iU

Implementation

The commits are organized by component to make the changes easier to review:

1. Write: Concurrent, batched UPSERT operations using withIdFn and withContentFn, with bounded in-flight concurrency and flushing in @FinishBundle.
2. JsonObjectCoder: UTF-8 JSON encoding for Couchbase SDK JsonObject values.
3. Read: Bounded SQL++ query execution as a single unit of work, with an optional parsing function (defaulting to JsonObject), named/positional parameters, scan consistency, and opt-in withRedistribute() applied after parsing.
4. Schema-aware reads: readRows(Schema) converts query results into Beam Row values using RowJson, respecting the supplied schema's nullability constraints.
5. Documentation: Class Javadocs, CHANGES.md, and the connector documentation table.
6. Review improvements: Display data, a build configuration comment, and an upsert idempotency test.

Testing

• 39 unit tests across CouchbaseIOTest and JsonObjectCoderTest.
• 15 Testcontainers integration tests across CouchbaseIOReadTest and CouchbaseIOWriteTest, using couchbase/server:enterprise-8.0.3.
• Tests cover authentication failures (the password must not appear in errors, toString, or display data), type mismatches, missing required fields, and repeated UPSERT operations.
• Local validation passes: ./gradlew :sdks:java:io:couchbase:build (Checkstyle, Javadoc, Checker Framework nullness validation, and dependency analysis).
• checkJavaLinkage could not run locally because of a corporate repository proxy restriction; CI validation is pending.
• TLS (couchbases://) is delegated to the Couchbase SDK but is not covered by an integration test.

Known limitations / follow-ups

• Connection lifecycle: One Couchbase Cluster per DoFn instance. JVM-wide sharing may be considered after benchmarking.
• Source parallelism: SQL++ reads execute as a single unit of work. Partitioned reads are deferred.
• SchemaTransform / Managed I/O: Deferred pending further community feedback (design doc §12, question 2).
• Additional capabilities: Advanced write modes, durability configuration, dead-letter handling, and DCP-based CDC are future work.

R: @kennknowles

Thank you for your contribution! Follow this checklist to help us incorporate your contribution quickly and easily:

• If this contribution is large, please file an Apache Individual Contributor License Agreement.
• Replace these instructions with a description of your change and why you have made it.
• Mention the appropriate issue in your description (for example: addresses #123 or fixes #456).
• Update CHANGES.md with noteworthy changes.

See CONTRIBUTING.md for more details.