Handle list content when stripping think blocks

[chaserobertsonn]

Summary

• flatten list-shaped assistant content before stripping think/tool-call blocks
• add regression coverage for content-part lists reaching strip_think_blocks

Verification

• ./venv/bin/python -m pytest tests/agent/test_message_content.py -q
• ./venv/bin/python -m pytest tests/agent/test_memory_provider.py::TestFlattenMessageContent -q

This fixes a production Donna failure where interim assistant delivery retried until API call #99 with: expected string or bytes-like object, got 'list'.

[alt-glitch]

This was generated by AI during triage.

Duplicate of #66218: both PRs add list-to-text coercion in strip_think_blocks before regex processing. #66218 is the earlier open implementation.

[kshitijk4poor]

Closing as duplicate — this bug is already fixed on main.

#66567 (by @teknium1, merged Jul 17) added the same list-content coercion at the exact same chokepoint in strip_think_blocks (agent/agent_runtime_helpers.py), with comprehensive block-type handling for thinking/text/output_text blocks. #66945 (by @kshitijk4poor, merged Jul 18) added belt-and-suspenders flatten_message_text() at the _interim_assistant_visible_text call site plus a traceback-based error classifier.

Your fix was correct and targeted the right spot. Thanks for the contribution!