fix: delegate Arrow schema conversion to PyIceberg

INSTALL-AND-REFERENCE.md

@@ -146,6 +146,15 @@ filters. Replace `pc.field("age") >= 18` with `table.column("age") >= 18`, and r
 `[a, b]` with `a & b`. Code using PyArrow for Arrow-side filtering can continue to use
 `pyarrow.compute.field()` outside the Tower table API.
 
+Arrow schemas passed to `create()` or `create_if_not_exists()` are validated directly by
+PyIceberg, which assigns field IDs and preserves nested nullability and `b"doc"` field
+metadata. Timestamp units from seconds through microseconds, UTC-zoned microsecond
+timestamps, `time64[us]`, `date32`, and Decimal128 values up to precision 38 are supported.
+Nanosecond timestamps are rejected by default instead of being silently downcast, as are
+`time32`, `time64[ns]`, `date64`, Float16, Decimal256, and non-UTC zoned timestamps. Convert
+those fields explicitly before creating the table when the loss is acceptable. PyIceberg's
+native validation exceptions propagate unchanged.
+
 ### dbt Core support
 
 ```bash

src/tower/_tables.py

@@ -39,7 +39,6 @@
 )
 from .exceptions import PyArrowFilterMigrationError
 from .tower_api_client.models import CatalogCredentials
-from .utils.pyarrow import convert_pyarrow_schema
 from .utils.tables import (
     make_table_name,
     namespace_or_default,
@@ -650,7 +649,9 @@ def create(self, schema: pa.Schema) -> Table:
 
         Args:
             schema (pa.Schema): The PyArrow schema defining the structure of the table.
-                This will be converted to an Iceberg schema internally.
+                PyIceberg validates it and assigns Iceberg field IDs. Lossy or
+                unsupported types, including nanosecond timestamps by default, are
+                rejected.
 
         Returns:
             Table: A new Table instance wrapping the created Iceberg table.
@@ -685,7 +686,7 @@ def create(self, schema: pa.Schema) -> Table:
         # along the way.
         table = catalog.create_table(
             identifier=table_name,
-            schema=convert_pyarrow_schema(schema),
+            schema=schema,
         )
 
         return Table(
@@ -712,8 +713,10 @@ def create_if_not_exists(self, schema: pa.Schema) -> Table:
 
         Args:
             schema (pa.Schema): The PyArrow schema defining the structure of the table.
-                This will be converted to an Iceberg schema internally. Note that this
-                schema is only used if the table needs to be created.
+                PyIceberg validates it and assigns Iceberg field IDs. Lossy or
+                unsupported types, including nanosecond timestamps by default, are
+                rejected. This schema is only used if the
+                table needs to be created.
 
         Returns:
             Table: A Table instance wrapping either the newly created or existing Iceberg table.
@@ -747,7 +750,7 @@ def create_if_not_exists(self, schema: pa.Schema) -> Table:
         # exists.
         table = catalog.create_table_if_not_exists(
             identifier=table_name,
-            schema=convert_pyarrow_schema(schema),
+            schema=schema,
         )
 
         return Table(