[Feature]: Custom subtype for step, callback, and child context operations #749

What would you like?

Let a caller set a custom subtype on a step, a callback, and a child context.

1. A function can compose several durable operations into one larger operation.
2. Every operation records a subtype in its checkpoint. Execution history and plugins read it.
3. In Python, StepConfig and CallbackConfig have no subtype field. ChildConfig.sub_type exists, but it is typed as the OperationSubType enum, and its docstring says the execution engine uses it internally.
4. So the parts of a composed operation all record Step, Callback, or RunInChildContext. History and plugins can tell them apart only by parsing their names.

The JS SDK supports this today: ChildConfig.subType, and since aws/aws-durable-execution-sdk-js#939, StepConfig.subType and CreateCallbackConfig.subType. The Go SDK has WithChildSubType.

Possible Implementation

1. Add sub_type: str | None = None to StepConfig and CallbackConfig. Widen ChildConfig.sub_type to accept a str as well as an OperationSubType.
2. Three places assume that every subtype is an enum value. Each would fail on a custom subtype:
   • OperationType.from_sub_type raises ValueError for an unknown subtype. OperationIdentifier uses it to derive the operation type. So the identifier needs the operation type passed in.
   • Operation.from_dict in lambda_service.py parses SubType with OperationSubType(...). An execution that recorded a custom subtype then fails to load its own state on the next invocation. So parsing must keep an unknown subtype as a string.
   • Serialization writes self.sub_type.value. It must also accept a string.
3. Use the custom subtype in checkpoint updates, replay validation, and plugin events.
4. Validate the value before the operation is created:
   • None or an empty string selects the default.
   • The service accepts 1 to 32 characters from [a-zA-Z0-9-_]. It rejects any other subtype with HTTP 400 ValidationException, and the execution then fails. So the SDK should reject such a value with a clear error.
   • Optionally reject the values the SDK uses for its own operations, as the Go SDK does.
5. Optionally, make the local checkpoint server in aws-durable-execution-sdk-python-testing reject an invalid subtype with the service's message. It already rejects a subtype that changes between updates of one operation. The JS testing SDK does this since feat(sdk): add subType to step and createCallback aws-durable-execution-sdk-js#939.

Nothing in the SDK decides how to suspend or retry from the subtype. So no other logic needs to change.

Is this a breaking change?

No. Without the option, the SDK records the same subtypes as today. Code that passes an OperationSubType to ChildConfig.sub_type keeps working.

Does this require an RFC?

No.

Additional Context

• Replay compares the subtype with the checkpoint. So changing a subtype is a breaking change for executions in flight. The docstrings should say this.
• Step 2 is a prerequisite. Without it, the first custom subtype makes the next invocation fail while it loads state.