Python: Fix handoff workflow with store=False sending server-assigned item IDs

[LEDazzio01]

Motivation and Context

Fixes #4357

When store=False, server-assigned item IDs (rs_* for reasoning, fc_* for function calls) reference non-existent server-persisted objects. During handoff workflows, these IDs are replayed in the input, causing Item not found API errors:

openai.BadRequestError: Error code: 400 - {
    'error': {
        'message': "Item 'rs_...' not found in the session's conversation.",
        'type': 'invalid_request_error',
        ...
    }
}

Description

This PR adds a post-processing step in _prepare_options() that strips the id field from reasoning and function_call input items when store=False. This ensures these items are replayed by value rather than by reference when server-side persistence is disabled.

Changes

_responses_client.py

• Added ID-stripping logic at the end of _prepare_options(), right before returning run_options
• Only activates when store is explicitly False
• Targets only reasoning and function_call item types

test_openai_responses_client.py

• Added test_prepare_options_strips_reasoning_and_function_call_ids_when_store_false — verifies IDs are removed from reasoning and function_call items when store=False
• Added test_prepare_options_preserves_reasoning_and_function_call_ids_when_store_true — verifies IDs are preserved when store=True

Both tests simulate a realistic handoff conversation with reasoning + function_call + function_result + text messages.

Contribution Checklist

• The code builds clean without any errors or warnings
• The PR title follows the repo convention
• Tests added to cover the change
• All new and existing tests pass

[Copilot]

Pull request overview

Fixes handoff workflows when store=False by ensuring OpenAI Responses API input items are replayed by value (not by reference) to avoid “Item not found” errors caused by server-assigned item IDs (rs_*, fc_*) that were never persisted.

Changes:

• Added a post-processing step in OpenAIResponsesClient._prepare_options() to remove id from reasoning and function_call input items when store is explicitly False.
• Added unit tests validating that these IDs are stripped for store=False and preserved for store=True.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

File	Description
python/packages/core/agent_framework/openai/_responses_client.py	Strips id fields from reasoning / function_call input items when store=False to prevent invalid references during replay.
python/packages/core/tests/openai/test_openai_responses_client.py	Adds coverage for ID stripping behavior under store=False and preservation under store=True.

Comments suppressed due to low confidence (1)

python/packages/core/tests/openai/test_openai_responses_client.py:2886

• The file now introduces a third # region ... marker but still has only a single # endregion at the end of the file, which makes region folding ambiguous (the Store=False section ends up nested under the Background Response section). Consider adding a matching # endregion for this new region (and/or closing the previous region before opening this one) to keep region markers balanced like other test files.

# region Store=False ID Stripping Tests (Issue #4357)

[TaoChenOSU]

nit: thank you for your contributions! We don't need to mention handoff here, which is likely to create more confusion.

[LEDazzio01]

Good point! Pushed commit eef02f1 — removed "handoff" mentions from the comment in _responses_client.py and the test docstring/comment in test_openai_responses_client.py. Now uses generic wording ("causing 'Item not found' API errors" and "Simulate a multi-turn conversation").

[moonbox3]

This works, but I'm worried about it living here as a post-processing step. If someone adds a new item type with a server-assigned ID down the road, they'd have to remember to update this loop too — and mcp_approval_request already has an id field that isn't covered here.

Could we thread store into _prepare_content_for_openai instead and just not emit the id at serialization time? Something like:

# reasoning case:
if content.id and store is not False:
    ret["id"] = content.id

That way each item type owns its own ID logic and new types get the right behavior by default.