[ShawnChen-Sirius]

Summary

• Close accepted Arrow readers on insert validation and native failures; preserve caller-owned batch roots.
• Add immutable named Arrow output choices while retaining the existing constructor.
• Bind typed named parameters in incremental Arrow queries through the core C API, reusing JDBC escaped-text encoding. Bump the JNI ABI to 4 to reject mismatched native packages at load time.
• Document ownership, parameter semantics, and the remaining batch-size and JDBC ResultSet integration limits.

Verification

• macOS arm64: rebuilt JNI shim against chDB core 26.11.0-rc.3 and ran mvn -pl chdb-arrow -am test with chdb.arrow.it=true.
• ArrowSessionTest: 29 tests, 0 failures, 1 skipped for the known core invalid String-to-Int64 insert regression. ArrowOptionsTest: 1 passed.

Follow-up to #38, which is merged into main. The local-engine build was only for testing; no local-engine package is intended for production.

Note

Add streaming named parameters to Arrow Java and fix reader ownership

• ArrowSession.queryArrow gains overloads that accept a named string parameter map; bindings are query-scoped and values are escaped via the JDBC text encoding (ArrowSession.java, ChdbConnection.java)
• ArrowOptions gets immutable named choices withUnsupportedAsBinary, withLowCardinalityAsDictionary, and withStringAsBinary that preserve other settings (ArrowOptions.java)
• New JNI entrypoint ChdbNative.arrowStreamOpenWithParams routes nonempty parameter maps to the core named-parameter API; empty maps use the existing path (chdb_jni.cpp)
• insertOwnedReader now takes ownership inside a try-with-resources after the open-session and no-active-query checks, so the reader closes even when table-name or schema validation fails (ArrowSession.java)
• Behavioral Change: JNI ABI version bumps from 3 to 4 in ChdbNative.JNI_ABI_VERSION, kJniAbiVersion, and the chdb.jni.abi.version Maven property; native shims and Java must be rebuilt together

^{Macroscope summarized d28b9f7.}