Implement custom serializers for pause/resume checkpoints

[rajkumar42]

Summary

Add support for custom serializers in the checkpoint/pause-resume system. This allows users to define how their custom objects are serialized and deserialized during agent execution pauses.

Context

The current checkpoint system (src/opensymbolicai/checkpoint.py) handles pause/resume for distributed execution. However, it may not gracefully handle custom object types that users define in their agents.

Expected Behavior

Users should be able to register custom serializers:

from opensymbolicai import register_serializer

@register_serializer(MyCustomClass)
class MyCustomSerializer:
    def serialize(self, obj: MyCustomClass) -> dict:
        return {"field": obj.field}
    
    def deserialize(self, data: dict) -> MyCustomClass:
        return MyCustomClass(field=data["field"])

Implementation Hints

• Add a serializer registry in checkpoint.py
• Provide a decorator or registration function
• Fall back to default JSON/pickle behavior if no custom serializer is found
• Consider using Pydantic's serialization capabilities

Acceptance Criteria

• Implement serializer registry
• Add @register_serializer decorator or equivalent API
• Add unit tests for custom serialization round-trips
• Add integration test with a real pause/resume cycle
• Document usage in docstrings

Files to Look At

• src/opensymbolicai/checkpoint.py — checkpoint/resume logic
• src/opensymbolicai/models.py — Pydantic models for reference

[Tanmay-Ts]

Hi, I’d like to work on this issue if it’s still relevant. Please let me know if that’s okay. Thanks!

[rajkumar42]

@Tanmay-Ts go for it!