Fix ClassCastException when flink read map and array data from parquet format data file.

[zhongyujiang]

This PR fixs ClassCastException when using Flink to read map and array data from parquet format data file. It adds a buildGenericMapData() in FlinkParquetReaders$ReusableMapData and adds a buildGenericArrayData() in FlinkParquetReaders$ReusableArrayData, and the XXReaders would build GenericXXData for return, which can be recognized by TypeSerializer of Flink.
Related to #3080.

[zhongyujiang]

@openinx @chenjunjiedada, could you help to review this PR? Thanks.

[coolderli]

Make sense to me!

[zlzhang0122]

LGTM！

[openinx]

In iceberg, we usually don't use * to import package, it's more clear to import the specify package one by one.

[openinx]

Thanks for reporting this bug, @zhong-yj ! The root cause is: apache flink couldn't serialize & deserialize customized MapData and ArrayData, technically it's a bug from apache flink , not iceberg bug. It's reasonable to customize our own ReusableArrayData to reuse the array data for saving memory purpose, and we should't copy all the elements from reusable array data for just fixing the serialize issue.

Actually, we apache flink 1.13 has a fix for this issue: https://issues.apache.org/jira/browse/FLINK-21247 . I think you can try to upgrade the flink to 1.13 to reproduce this bug .

Though we've a fix in flink 1.13, but it still expose an import issue in our apache iceberg flink module: we don't have SQL nested data unit tests to address this. I think it's worth to public PR to address this testing thing ( that may need us to merge the #2629 firstly).

[openinx]

I tried this by building flink-iceberg-runtime.jar and running the SQL in flink 1.13.2 local cluster. All things works as expected
.

Let's just close this PR , and turn to upgrade the flink version to 1.13.2 #2629 . Close this now. Thanks @zhong-yj for your reporting and contribution !

[openinx]

This bug has been addressed since we've successfully integrated to flink 1.13.2 in here : #3116