docs: Add Python session identity ADR

[eavanvalkenburg]

Motivation & Context

Python protocol and service integrations currently use session, task, response, and continuation identifiers with overlapping terminology. This ADR documents the lifetime boundaries for those identities so follow-up implementation work for native conversation propagation and A2A task/session behavior has a clearer design basis.

The ADR keeps AG-UI out of the session identity decision, preserves service_session_id as an opaque service continuation handle, and focuses the remaining open design question on how Python should store additional durable continuation state across calls.

Description & Review Guide

• What are the major changes?
  • Adds ADR 0029 for Python identity lifetimes across sessions, tasks, service continuation, response IDs, and continuation tokens.
  • Records the chosen direction to split identity storage by lifecycle.
  • Lists the remaining durable-state options: protocol-specific AgentSession subclasses, namespaced AgentSession.state, or richer service-owned service_session_id values.
  • Adds an A2A appendix for task_id vs reference_task_ids behavior to validate in implementation.
• What is the impact of these changes?
  • No runtime behavior changes.
  • Provides design guidance for future Python work on Python: [Bug]: No Native Conversation ID Propagation #4673 and Python: [Feature]: Align concepts between AG-UI and A2A protocols #4893.
• What do you want reviewers to focus on?
  • Whether the ADR correctly separates service continuation, durable protocol state, response identity, and continuation tokens.
  • Whether the remaining durable-state options are framed clearly enough for a follow-up decision.

Related Issue

Related to #4673 and #4893. Does not close them.

Contribution Checklist

• The code builds clean without any errors or warnings
• All unit tests pass, and I have added new tests where possible
• The PR follows the Contribution Guidelines
• This PR is linked to an issue and there is no other open PR for this issue (see Related Issue above).
• This is not a breaking change. If it is a breaking change, add the breaking change label (or add "[BREAKING]" to the title prefix, before or after any language prefix) — a workflow keeps the label and title prefix in sync automatically.

[github-actions]

Automated Code Review

Reviewers: 5 | Confidence: 85% | Result: All clear

Reviewed: Correctness, Security Reliability, Test Coverage, Failure Modes, Design Approach

---

Automated review by eavanvalkenburg's agents

[Copilot]

Pull request overview

Adds ADR 0029 to document Python identity lifetimes and boundaries across AgentSession, protocol task/session identifiers, response/message IDs, and framework continuation tokens, to reduce terminology overlap and guide follow-up work for native conversation propagation and A2A behavior.

Changes:

• Introduces ADR 0029 describing identity lifetimes and where each identifier should live (durable session state vs. per-response vs. continuation token vs. run/telemetry correlation).
• Documents current Python implementation behavior for A2A and OpenAI Responses continuation to ground the decision discussion.
• Enumerates remaining durable-state shape options (session subclasses vs. namespaced session.state vs. richer service_session_id) plus an A2A task_id/reference_task_ids appendix.

[github-actions]

Automated Code Review

Reviewers: 5 | Confidence: 84% | Result: All clear

Reviewed: Correctness, Security Reliability, Test Coverage, Failure Modes, Design Approach

---

Automated review by eavanvalkenburg's agents

[TaoChenOSU]

This sounds like the gap that exists today is primarily in A2A. Is it necessary to introduce a new abstraction for me? My understanding of the problem is very limited. Please correct me if I am wrong.

[eavanvalkenburg]

@TaoChenOSU Thanks — your review question on whether this needs a new abstraction is now reflected directly in the Decision section. The direction is to avoid a new top-level identity abstraction and keep continuation identity under service_session_id (Option B), while keeping run correlation in existing run/telemetry context. The immediate implementation gap is mostly A2A mapping clarity, but we keep the lifecycle split consistent across providers.