[Feature]: Expand SerDesContext with payload and operation metadata

## What would you like?

Expand `SerDesContext` beyond `operation_id` and `durable_execution_arn` so context-aware SerDes implementations can identify the exact payload being processed.

Useful metadata includes:

- stable entity ID;
- payload kind, such as execution input/output, operation result, invoke payload, state, or exception;
- operation name, parent ID, type, and subtype;
- retry attempt number when applicable.

This enables safer external-storage ownership rules, better diagnostics, and stable naming without parsing operation IDs.

## Possible Implementation

- Add optional fields to the frozen `SerDesContext` dataclass while preserving `operation_id` and `durable_execution_arn`.
- Populate the fields for every SDK-managed serialization and deserialization path, including execution input/output, steps, child contexts, callbacks, wait-for-condition state, invoke payload/results, exceptions, and retry attempts.
- Document which fields are always present and which are operation-specific.
- Ensure the same context is passed to every attempt of a retryable SerDes operation.
- Add parity tests that capture context from each payload path and validate first execution versus replay.
- Use the richer context when implementing owner-bound filesystem envelopes and safely rebinding forwarded exception data.

## Is this a breaking change?

No. New dataclass fields should have defaults so existing custom SerDes implementations and direct context construction continue to work.

## Does this require an RFC?

Yes.

## Additional Context

The Java SDK work in aws/aws-durable-execution-sdk-java#648 passes payload kind, stable entity identity, operation metadata, and attempt number explicitly to SerDes stages.

Commencez par localiser la dataclass gelée SerDesContext et les chemins de sérialisation et de désérialisation gérés par le SDK listés dans l’issue. Examinez comment l’input/output d’exécution, les étapes, les callbacks, les payloads d’invoke, les exceptions et les retries transmettent actuellement le contexte. C’est terminé lorsque les champs de métadonnées facultatifs sont documentés et renseignés de manière cohérente, avec des tests de parité couvrant chaque chemin de payload et le comportement de replay.

Rédigé par le modèle d'indexation à partir du texte de l'issue.