fix: sanitize orphan tool results in continue history

[ki113r4bbi7]

What

• Sanitize native /continue history before restoring it.
• Preserve orphan tool_result content as plain text instead of replaying it as a protocol tool result.
• Keep valid tool_use / tool_result pairs unchanged.

Why

If an agent crashes while a tool call is in flight, a recovered session can contain a tool_result without the matching prior assistant tool_use. Replaying that history can make providers reject the next request with errors such as No tool call found for function_call_output.

Testing

• python -m py_compile frontends/continue_cmd.py
• Focused import test for _parse_native_history:
  • orphan tool_result is converted to text
  • matching tool_result remains intact