[Bug]: wait_for_callback errors lose callback_id during reconstruction

[zhongkechen]

Expected Behavior

Callback errors surfaced by DurableContext.wait_for_callback should retain the callback ID generated for the operation, consistent with the CallbackError.callback_id field and the callback error hierarchy introduced by #563. The value should be available both when the error first surfaces and on replay.

Actual Behavior

CallbackExternalError and CallbackTimeoutError are reconstructed at the failed child-context boundary with callback_id=None. The original nested __cause__ is also lost. CallbackSubmitterError is constructed without a callback ID as well.

The direct Callback.result() error is initially created with the ID, but ChildOperationExecutor serializes it through ErrorObject. DurableOperationError.from_error_fields() then reconstructs the registered callback type using only message, error_type, data, and stack_trace; callback_id defaults to None. wait_for_callback subsequently re-raises that reconstructed exception.

Steps to Reproduce

1. Define a durable handler that calls context.wait_for_callback(...) and catches CallbackError.
2. Start the execution and obtain the callback ID from the callback operation.
3. Complete the callback with SendDurableExecutionCallbackFailure or the local runner equivalent.
4. Read error.callback_id in the handler.

Expected: the ID sent to the external caller.

Actual: None.

A local durable-runner regression test against PR #563 reproduced:

CallbackExternalError.callback_id == None
CallbackExternalError.__cause__ == None

SDK Version

Unreleased PR #563 (2ac05c8d712bcdac492eee07ddc930f7f9ef2088)

Python Version

3.14

Is this a regression?

No. Before #563, wait_for_callback surfaced a ChildContextError and the callback ID was already unavailable. This issue tracks completing the new callback-error metadata contract rather than restoring previously working behavior.

Additional Context

Relevant paths:

• ErrorObject.raise_as_operation_error() reconstructs the inner exception.
• DurableOperationError.from_error_fields() does not pass callback_id to callback subclasses.
• DurableContext.wait_for_callback() re-raises the reconstructed callback error and constructs CallbackSubmitterError without an ID.

Add a durable-runner test that catches each callback error type across the actual child-context checkpoint/replay boundary; the existing mocked pass-through test does not exercise serialization.

[ayushiahjolia]

callback_id is not carried on callback errors. It isn't an error wire field, so it can't survive the wait_for_callback child-context boundary, and once that context is terminal (replay_children=false) its child operations are dropped from later invocations' InitialExecutionState - the CALLBACK op, and any id derived from it, is absent on replay. Rather than expose a field that is populated on first run and None after replay, we omit it; the caller already holds it from create_callback() / the submitter argument.