[codemod][py3.12] Convert dataclasses mutable default values to use default_factory

Summary:
As of [Python 3.11](https://docs.python.org/3.11/whatsnew/3.11.html#dataclasses), dataclasses now only allow defaults that are hashable. Using mutable (non-hashable) default values under Python 3.11+ will cause a runtime error:

```
ValueError: mutable default <class 'problematic.ClassName'> for field <field> is not allowed: use default_factory
```

This codemod attempts to automatically convert mutable defaults to use `default_factory`

NOTE: The change is not semantically equivalent to before the change. Before, all dataclass instances with a mutable default value were sharing the same instance. This change results each dataclass instance using a new instance of the mutable value. It is likely that the before state was a latent bug, but it's still a behavior change!

Reviewed By: moutaigua8183

Differential Revision: D73523400

fbshipit-source-id: c00bb2dc5bcf1df22ce6a663588fc2d5dd1a36be

Author: generatedunixname647790274085263

mmf/models/alignment.py

@@ -2,7 +2,7 @@
 
 import math
 from copy import deepcopy
-from dataclasses import dataclass
+from dataclasses import dataclass, field
 from typing import Any, Tuple
 
 import torch
@@ -82,8 +82,8 @@ class Config(BaseModel.Config):
         direct_features_input: bool = False
         image_encoder: Any = MISSING
         text_encoder: Any = MISSING
-        image_projection: Any = IdentityEncoder.Config()
-        text_projection: Any = IdentityEncoder.Config()
+        image_projection: Any = field(default_factory=lambda: IdentityEncoder.Config())
+        text_projection: Any = field(default_factory=lambda: IdentityEncoder.Config())
 
     def __init__(self, config: Config):
         """Initialize the config which is the model configuration."""