chore: remove dead native_iceberg_compat JVM Parquet reader chain

Now that native_iceberg_compat is gone, the JVM-mediated V1 Parquet read
path (CometScanExec → CometScanWrapper → CometParquetFileFormat →
NativeBatchReader / column-level Native APIs) is unreachable. Iceberg
also no longer integrates against Comet's @IcebergApi surface.

- CometExecRule: V1 CometScanExec fallback now reverts to wrapped
  FileSourceScanExec (Spark) instead of CometScanWrapper.
- CometScanExec: keep as planning intermediate, throw on
  doExecute/doExecuteColumnar/executeCollect/inputRDDs.
- Drop file-format swap to CometParquetFileFormat in both
  CometScanExec.apply and CometNativeScanExec.apply.
- Delete CometParquetFileFormat and the JVM Parquet reader chain
  (NativeBatchReader, NativeColumnReader, ColumnReader, RowGroupReader,
  FileReader, plus all supporting page/index/options/utility classes).
- Delete IcebergCometNativeBatchReader and CometLazyVector.
- Trim Native.java to the Arrow-native scan APIs.
- Delete IcebergApi annotation, CometSchemaImporter,
  AbstractCometSchemaImporter; strip @IcebergApi from CometVector.
- Delete now-orphaned tests (TestFileReader, TestUtils,
  TestCometInputFile); drop closeColumnReader NPE assertion in
  CometNativeSuite.

Author: andygrove