feat(parquet): expose ParquetReadOptions via registerParquet and readParquet

[andygrove]

Closes #17.

Exposes DataFusion's ParquetReadOptions on the Java surface and adds SessionContext.readParquet(...).

What's in this PR

• org.apache.datafusion.ParquetReadOptions — mutable fluent class with five setters: fileExtension, parquetPruning, skipMetadata, metadataSizeHint, schema (Arrow Java Schema).
• SessionContext.registerParquet(name, path, options) — overload of the existing method. Existing registerParquet(name, path) is preserved (delegates to the overload with default options).
• SessionContext.readParquet(path[, options]) -> DataFrame — analog of Rust's ctx.read_parquet, returns a DataFrame without registering.
• Schema is marshalled across JNI as Arrow IPC stream bytes (reuses the IPC mechanism added in feat(proto): execute Java-built protobuf plans via SessionContext #13's tableSchema).
• 7 new tests: 3 unit tests for the options class, 4 integration tests (default options row count, custom file extension, explicit schema, metadata size hint).

Not in this PR

• table_partition_cols (Hive-style partition columns) — requires per-column ArrowType marshalling.
• file_sort_order — requires logical-Expr serialization.
• file_decryption_properties — encryption; niche.
• A static-factory builder() — fluent setters suffice.
• CSV/JSON/Avro analog option classes.