GeezerVison

Which issue does this PR close?

• Closes Add support for Java data sources #63.

Rationale for this change

Java users have no way to expose custom in-process tables (JDBC scans, in-memory
collections, custom file formats, etc.) to DataFusion. This adds a minimal
DataSource interface and the JNI wiring to register it on a SessionContext.
The implementation mirrors the existing scalar-UDF JNI pattern.

What changes are included in this PR?

• New public DataSource interface in org.apache.datafusion with
  Schema schema() and ArrowReader scan(BufferAllocator).
• SessionContext.registerDataSource(name, source) registers a Java-backed
  table; schema is captured at registration time.
• JniBridge.invokeDataSourceScan exports the user's ArrowReader through
  the Arrow C Data Interface (zero-copy).
• Native: JavaDataSource: TableProvider + JavaScanExec: ExecutionPlan in
  native/src/data_source.rs, plus the JNI entry point.
• Shared jthrowable_to_string helper lifted into native/src/jni_util.rs
  so the UDF and data-source paths share Java-exception formatting.
• New JdbcExample in the examples module demonstrating an end-to-end
  JDBC-backed DataSource: populates an H2 in-memory table, wraps a JDBC
  query in a JdbcDataSource, registers it, and runs an aggregation query.
  Streams batches via arrow-jdbc's ArrowVectorIterator wrapped in a small
  ArrowReader subclass — no IPC re-serialisation. Adds arrow-jdbc and
  com.h2database:h2 as examples-module deps.
• v1 scope: single partition, no projection or filter pushdown into Java
  (DataFusion projects/filters on top), no deregisterTable. Multi-partition,
  pushdown, and deregistration are listed as follow-ups in the user guide.

Run the JDBC example with:

./mvnw -pl examples exec:exec -Dexec.mainClass=org.apache.datafusion.examples.JdbcExample

Are these changes tested?

Yes — eight integration tests in
`core/src/test/java/org/apache/datafusion/DataSourceTest.java`:

• `SELECT *` happy path
• `UNION ALL` over the same registered table (multi-scan)
• Empty stream
• Column projection through DataFusion
• Two registered tables joinable in one query
• Schema-mismatch surfaces a readable error
• `scan()` throwing propagates the Java exception class and message
• `scan()` returning null is rejected with `IllegalStateException`

The JDBC example is exercised end-to-end manually (output verified: aggregation
produces `alice → 119.99` and `bob → 7.50` over the H2 fixture). It compiles
as part of the standard `mvn package` build alongside the other example
classes (`AddOneExample`, `DataFrameExample`, etc.) — none of which carry
JUnit tests, by convention.

Are there any user-facing changes?

Yes — new public `DataSource` interface and `SessionContext.registerDataSource`
method, plus a new user-guide page at `docs/source/user-guide/data-source.md`
covering the API, contract, threading, errors, and v1 limitations. The
runnable `JdbcExample` shows the API in action against an embedded H2.