Add design topic page on use of Python builtin types

spec/design_topics/index.rst

@@ -7,3 +7,4 @@ Design topics & constraints
 
    backwards_compatibility
    data_interchange
+   python_builtin_types

spec/design_topics/python_builtin_types.md

@@ -0,0 +1,41 @@
+# Python builtin types and duck typing
+
+Use of Python's builtin types - `bool`, `int`, `float`, `str`, `dict`, `list`,
+`tuple`, `datetime.datetime`, etc. - is often natural and convenient. However,
+it is also potentially problematic when trying to write performant dataframe
+library code or supporting devices other than CPU.
+
+This standard specifies the use of Python types in quite a few places, and uses
+them as type annotations. As a concrete example, consider the `mean` method and
+the `float` it is documented to return, in combination with the `__gt__` method
+(i.e., the `>` operator) on the dataframe:
+
+```python
+class DataFrame:
+    def __gt__(self, other: DataFrame | Scalar) -> DataFrame:
+        ...
+    def get_column_by_name(self, name: str, /) -> Column:
+        ...
+
+class Column:
+    def mean(self, skipna: bool = True) -> float:
+        ...
+
+larger = df2 > df1.get_column_by_name('foo').mean()
+```
+
+For a GPU dataframe library, it is desirable for all data to reside on the GPU,
+and not incur a performance penalty from synchronizing instances of Python
+builtin types to CPU. In the above example, the `.mean()` call returns a
+`float`. It is likely beneficial though to implement this as a library-specific
+scalar object which duck types with `float`. This means that it should (a) have
+the same semantics as a builtin `float` when used within a library, and (b)
+support usage as a `float` outside of the library (i.e., implement
+`__float__`). Duck typing is usually not perfect, for example `isinstance`
+usage on the float-like duck type will behave differently. Such explicit "type
+of object" checks don't have to be supported.
+
+The following design rule applies everywhere builtin Python types are used
+within this API standard: _where a Python builtin type is specified, an
+implementation may always replace it by an equivalent library-specific type
+that duck types with the Python builtin type._