Python: fix thread serialization for multi-turn tool calls

[Eduard van Valkenburg (eavanvalkenburg)]

Motivation and Context

Fixes #4672.

Persisted Python sessions can retain OpenAI Responses fc_id values on stored function_call content. Those IDs are response-scoped and valid only for the original Responses tool loop. The bug is real, but the fix belongs in the Responses request-preparation path rather than in framework-wide session loading.

Description

• move stale-fc_id replay handling out of _sessions.py and into OpenAIResponsesClient._prepare_messages_for_openai()
• preserve stored fc_id values in session history, but ignore them when replaying provider-attributed history back through the Responses API
• keep live same-run tool-loop fc_id behavior unchanged so active Responses tool calls still use their real item IDs
• add/update regressions for in-memory history reloads, session serialization, cross-provider shared local storage, and duplicate call_id reuse between replayed history and live tool calls

Contribution Checklist

• The code builds clean without any errors or warnings
• The PR follows the Contribution Guidelines
• All unit tests pass, and I have added new tests where possible
• Is this a breaking change? If yes, add "[BREAKING]" prefix to the title of the PR.

[Copilot]

Pull request overview

Fixes restored Python session replay for OpenAI Responses multi-turn tool calls by ensuring provider-ephemeral function-call IDs (fc_id) are not resent when history is reloaded into a new run context.

Changes:

• Strip fc_id from function_call content when loading stored history into SessionContext (without mutating stored session state).
• Add a regression test that covers in-memory history loading plus session serialization/deserialization round-trips for multi-turn tool calls.

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/_sessions.py	Sanitizes loaded history by removing provider-ephemeral fc_id before extending messages into the run context.
python/packages/core/tests/openai/test_openai_responses_client.py	Adds an async regression test verifying fc_id is preserved live but not replayed from loaded/restored history.

---

You can also share your feedback on Copilot code review. Take the survey.

[Mark Wallace (markwallace-microsoft)]

Python Test Coverage Report •

File	Stmts	Miss	Cover	Missing
packages/core/agent_framework/openai
_responses_client.py	821	126	84%	312–315, 319–320, 325–326, 336–337, 344, 359–365, 386, 394, 417, 514, 613, 672, 674, 676, 678, 746, 760, 840, 850, 855, 898, 977, 994, 1007, 1072, 1165, 1170, 1174–1176, 1180–1181, 1247, 1276, 1282, 1292, 1298, 1303, 1309, 1314–1315, 1376, 1398–1399, 1414–1415, 1433–1434, 1475–1478, 1640, 1695, 1697, 1777–1785, 1907, 1962, 1977, 1997–2007, 2020, 2031–2035, 2049, 2063–2074, 2083, 2115–2118, 2126–2127, 2129–2131, 2145–2147, 2157–2158, 2164, 2179
TOTAL	24006	2640	89%

Python Unit Test Overview

Tests	Skipped	Failures	Errors	Time
5234	20 💤	0 ❌	0 🔥	1m 23s ⏱️