fix: prune stale tool results during session repair

[he-yufeng]

Summary

• prune orphaned or duplicate toolResult blocks while repairing restored session histories
• keep the existing matching results and add only the missing interruption results, so the next user message matches the preceding assistant toolUse blocks
• add a regression test for mixed duplicate/stale/missing tool-result history

Fixes #2296

Tests

• .venv\Scripts\python.exe -m pytest tests\strands\session\test_repository_session_manager.py -q -k "fix_broken_tool_use"
• .venv\Scripts\python.exe -m pytest tests\strands\session\test_repository_session_manager.py -q
• $env:RUFF_CACHE_DIR='.tmp\ruff-cache'; .venv\Scripts\python.exe -m ruff check src\strands\session\repository_session_manager.py tests\strands\session\test_repository_session_manager.py
• $env:RUFF_CACHE_DIR='.tmp\ruff-cache'; .venv\Scripts\python.exe -m ruff format --check src\strands\session\repository_session_manager.py tests\strands\session\test_repository_session_manager.py
• .venv\Scripts\python.exe -m py_compile src\strands\session\repository_session_manager.py tests\strands\session\test_repository_session_manager.py
• git diff --check

[xiaosu19]

Thanks for the fix! We hit this in production with AgentCore Memory + GLM-5 on Bedrock. Verified that the root cause matches — our workaround was to prevent corrupted history at the application layer (BeforeToolCallEvent + cancel_tool), but this SDK-level fix is the proper solution.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!