Support PyArrow sorted MapType schemas

Problem

Lance currently rejects or loses Arrow MapType schemas with keys_sorted=true in several paths. This blocks round-tripping PyArrow schemas such as:

pa.map_(pa.string(), pa.int32(), keys_sorted=True)

The limitation was originally called out in code comments around map support because Rust/PyArrow schema conversion could drop the keys_sorted property.

Expected behavior

Lance should preserve MapType.keys_sorted for V2.2 map data across write and read paths, including:

• dataset writes and reads
• scanner/table/batch reads
• fragment writes and reads
• direct file reader APIs
• Lance schema conversion APIs
• namespace schema JSON conversion

Notes

The implementation should keep normal unsorted maps unchanged and avoid adding per-batch Python wrapping for schemas that do not need sorted-map restoration.

Start by tracing the Rust/PyArrow schema conversion paths and the dataset, scanner/table/batch, fragment, direct file reader, Lance schema, and namespace JSON APIs named in the issue. Verify how V2.2 MapType schemas are handled for sorted and unsorted maps across writes and reads. Done means keys_sorted=true round-trips through every listed path without changing normal maps or adding unnecessary per-batch Python wrapping.