fix(template): replace rejected responses after the last user message

[Excelius-Wang]

Agent preference data using tool messages currently inherits chosen tool calls and results when building rejected_messages from rejected_response. Keep the documented replacement boundary at the last user message by adding an opt-out for tool boundaries to the existing helper. Its default behavior, including the last-round loss fix in #9923, remains unchanged.

Validation: 10 tests and 12 subtests passed, including an actual JSONL/OpenAI tool-call loading regression and the existing loss-scale tests. Local CPU encoding checks covered 48 official-example variants and 192 cases from three public preference datasets; all match explicit rejected-message references after the fix. Six two-step tiny-model DPO runs produced identical losses and trained weights for compact versus explicit representations on each dataset. Changed-file pre-commit checks and git diff --check passed.

[tastelikefeet]

The documentation need to change.
to declare that, the rejected_response needs to contain each rejected round after the last user field, including tool round, else user will not know to add negative tool rounds.

[Excelius-Wang]

Updated the Chinese and English docs and added an example showing how to include tool calls and results in rejected_response. Thanks for pointing this out!