fix(java): add diagnostics for imported Arrow streams

[beinan]

Summary

Add Lance Java diagnostics around Arrow C Data stream readers and a scanner regression for the nested struct case that motivated the lance-spark fix.

Findings

The original VectorLoader failure can be caused by consumers closing or mutating the reader-owned VectorSchemaRoot between loadNextBatch() calls. When the damaged vector is a StructVector, Arrow Java may see schema children still present while the reused vector has zero children, producing should have as many children as in the schema: found 0 expected N.

Changes

• Add LanceArrowReaders, a small wrapper for Arrow readers imported from Lance Arrow C Data streams.
• Use the wrapper for Java APIs that call Data.importArrayStream(...), including LanceScanner, AsyncScanner, SQL query output, delta output, and file reads.
• Preserve the underlying reader behavior while adding Lance context to batch-load failures, including the child-count mismatch diagnostic.
• Add a Java scanner regression that writes a six-field all-null struct and scans it across multiple batches.
• Add a unit test for the diagnostic message.

Testing

cd java
./mvnw -Dtest=org.lance.ScannerTest#testDatasetScannerAllNullStructColumnMultipleBatches,org.lance.ipc.LanceArrowReadersTest test

This run passed Java checkstyle, Spotless, both targeted Java tests, the JNI build, and the rust-maven-plugin JNI test.

[claude]

Claude Code Review

This pull request is from a fork — automated review is disabled. A repository maintainer can comment @claude review to run a one-time review.