Fix dataclass field types (#257)

There is a (Python?) bug where calling
`dataclasses.fields(some_data_class)` sometimes returns fields where
`type` is a type _name_ instead of a _type_ (for example `'str'` instead
of `<class str>`). This prevents our ORM from working as expected.
The issue can be reproduced in
databrickslabs/ucx#2401, and is currently fixed
using a local workaround see
https://github.com/databrickslabs/ucx/pull/2401/files#diff-40a8f3f4a3a5e5b222ccbfb95204001a6b007c78cbe55b9f1a53af5a8309886eR349
This PR works around the issue globally.

---------

Co-authored-by: Eric Vergnaud <eric.vergnaud@databricks.com>

Author: ericvergnaud

src/databricks/labs/lsql/backends.py

@@ -70,6 +70,13 @@ def _schema_for(cls, klass: Dataclass):
         fields = []
         for f in dataclasses.fields(klass):
             field_type = f.type
+            # workaround rare (Python?) issue where f.type is the type name instead of the type itself
+            # this seems to happen when the dataclass is first used from a file importing it
+            if isinstance(field_type, str):
+                try:
+                    field_type = __builtins__[field_type]
+                except TypeError as e:
+                    logger.warning(f"Could not load type {field_type}", exc_info=e)
             if isinstance(field_type, UnionType):
                 field_type = field_type.__args__[0]
             if field_type not in cls._builtin_type_mapping: