Add native subclasses for InputFile and OutputFile to simplify.

[rdblue]

This is an update to apache#9359 that simplifies the read path and addresses review items.

This introduces NativeEncryptionInputFile to solve problems in the read path. Specifically the encryption manager needed to know whether to decrypt or return the underlying input file, but that depends on the file format and whether that format is capable of native encryption. In addition, the read path also needed to know whether to configure native encryption, which was a separate choice. And I think having a separate encryption choice would also have broken existing encryption managers that handle Parquet files with encrypting streams.

Instead, this uses a new InputFile type to signal that native encryption should be used if possible and to carry the encryption key. Then the format helpers (Avro, Parquet, and ORC) are responsible for either using the encrypting stream (Avro) or applying native encryption (Parquet and ORC).

I considered other approaches here, but this was the cleanest because most others required separate handling for encryption keys and didn't cover all paths -- notably missing the Arrow readers. By using the same InputFile everywhere, this requires fewer changes to Spark and Flink readers and should work everywhere that the encryption manager is used.

[rdblue]

This was needed to avoid making StandardKeyMetadata public.