REF: dtypes.cast

[jbrockmendel]

There are too many functions in dtypes.cast, making it difficult to know which one to use in a given situation. e.g. when working with Index.insert in #38102 I considered maybe_promote, maybe_upcast, maybe_casted_values, find_common_type.

Best case scenario we can de-duplicate some of these and just remove some of them, but failing that, I think we could find new homes for e.g. construct_1d_arraylike_from_scalar, construct_1d_object_array_from_listlike, construct_1d_ndarray_preserving_na.

[jreback]

would start by making a submodule and organizing

[jbrockmendel]

There are a couple of functions in here that are basically disguised Series/DataFrame methods, in particular cast.convert_dtypes is effectively Series._convert_dtypes, cast.maybe_casted_values is only used in DataFrame.reset_index. These aren't re-usable low-level functions (which is how i like to think of core.dtypes, but I guess it doesn't have to be that way), but I'm not eager to move code into Series/DataFrame, so not obvious how to handle.