Docs: Add missing _downcast_ns_timestamp_to_us to ArrowScan docstring

[qzyu999]

Which issue does this PR close?

Minor documentation gap (no issue).

Rationale for this change

The ArrowScan class declares _downcast_ns_timestamp_to_us: bool | None as a class-level type annotation and documents all other attributes in the docstring, but this attribute was missing from the Attributes section.

What changes are included in this PR?

Adds one line to the ArrowScan class docstring documenting the _downcast_ns_timestamp_to_us attribute.

Are there any user-facing changes?

No.

[ebyhr]

Looks good to me. The constant at L1767 contains "ON_WRITE", but the actual usage is on read in my understanding.

[qzyu999]

Looks good to me. The constant at L1767 contains "ON_WRITE", but the actual usage is on read in my understanding.

Hi @ebyhr, good catch on DOWNCAST_NS_TIMESTAMP_TO_US_ON_WRITE. It does seem to be used for writing in /pyiceberg/table/__init__.py (append(), overwrite(), upsert(), add_files()) and pyiceberg/io/pyarrow.py (write_file(), _dataframe_to_data_files()), while using it for schema inference in pyiceberg/catalog/__init__.py (_convert_schema_if_needed). It's really only in this ArrowScan where it's quite misleading.

I can create a separate PR to rename it to DOWNCAST_NS_TIMESTAMP_TO_US while keeping it backwards compatible since it's a user-facing variable, WDYT?

Edit: Opened a new PR (#3596) for renaming DOWNCAST_NS_TIMESTAMP_TO_US_ON_WRITE.

[Fokko]

Thanks for adding this @qzyu999