The current SessionContext.registerParquet(name, path) is hardwired to ParquetReadOptions::default(), and there's no Java entry point for DataFusion's read_parquet(path, options) -> DataFrame.

Expose:

• A ParquetReadOptions Java class with fluent setters for fileExtension, parquetPruning, skipMetadata, metadataSizeHint, and an explicit schema (Arrow Java Schema).
• An overload SessionContext.registerParquet(name, path, options).
• A new method SessionContext.readParquet(path[, options]) that returns a DataFrame without registering.

Marshalled across JNI as flat primitive arguments plus optional Arrow IPC bytes for the schema (reuses the IPC mechanism from #8 / #13).

Out of scope for this issue: table_partition_cols, file_sort_order, file_decryption_properties. CSV/JSON/Avro analog option classes.