[python][daft] Make Daft Paimon write sink serializable

[kerwin-zk]

Purpose

PaimonDataSink currently keeps the FileStoreTable and WriteBuilder directly in the sink object. When Daft runs with the Ray runner, the sink needs to be serialized and sent to workers. For OSS/Jindo tables, the table can indirectly hold PyArrow/Jindo filesystem objects that are not picklable, causing Ray serialization failures.

  Checking PaimonDataSink daft.pickle roundtrip ...
  Traceback (most recent call last):
    File ".../test_paimon_daft_filesystem.py", line 72, in <module>
      restored_sink = loads(dumps(PaimonDataSink(table, mode="overwrite")))
                            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  ...
    File "<stringsource>", line 2, in _fs.FileSystem.__reduce_cython__
  TypeError: no default __reduce__ due to non-trivial __cinit__

This PR makes the Daft Paimon write sink serializable by only storing reconstructable table state during pickling.

Tests

CI

[leaves12138]

LGTM.

I reviewed the pickle/reconstruction path for PaimonDataSink. The state now avoids carrying the table/FileIO objects, reconstructs the table from catalog options + identifier (or table path fallback), and preserves the commit user / overwrite mode when rebuilding the write builder. The changed files compile and the PR checks are green.