Fix serve tool parsing for legacy response schemas

[Sunt-ing]

What does this PR do?

#46980 made new-style response templates require prefix=, but legacy response_schema rejects prefix. parse_tool_calls currently passes prefix="" unconditionally, so the serving tool-call path crashes for legacy schemas and the Qwen3.5 fallback schema before parsing any tool call.

This only passes prefix="" for new-style schemas (version or fields) and keeps legacy schemas on the no-prefix path. The existing new-style fallback remains covered, and this adds a regression guard for the legacy fallback.

Repro and output

from transformers.cli.serving.utils import _TOOL_CALL_FALLBACKS, parse_tool_calls
from transformers.tokenization_utils_base import PreTrainedTokenizerBase


class Processor:
    parse_response = PreTrainedTokenizerBase.parse_response


processor = Processor()
schema = next(v["schema"] for k, v in _TOOL_CALL_FALLBACKS.items() if "qwen3_5" in k)
text = "<function=get_weather><parameter=city>Paris</parameter></function>"

print("schema keys", sorted(schema.keys()))
print("text", text)
print(parse_tool_calls(processor, text, schema))

Before this PR:

schema keys ['items', 'type', 'x-regex-iterator']
text <function=get_weather><parameter=city>Paris</parameter></function>
ValueError: `prefix=` is only supported with new-style `response_template` specs, not legacy `response_schema`.

After this PR:

schema keys ['items', 'type', 'x-regex-iterator']
text <function=get_weather><parameter=city>Paris</parameter></function>
[{'name': 'get_weather', 'arguments': '{"city": "Paris"}'}]

Tests

python -m pytest -q tests/cli/test_serve.py::TestToolCallUnit::test_parse_tool_calls_from_text tests/cli/test_serve.py::TestToolCallUnit::test_parse_multiple_tool_calls_from_text tests/cli/test_serve.py::TestToolCallUnit::test_parse_tool_calls_with_legacy_schema_does_not_pass_prefix

tests/cli/test_serve.py::TestToolCallUnit::test_parse_tool_calls_from_text PASSED [ 33%]
tests/cli/test_serve.py::TestToolCallUnit::test_parse_multiple_tool_calls_from_text PASSED [ 66%]
tests/cli/test_serve.py::TestToolCallUnit::test_parse_tool_calls_with_legacy_schema_does_not_pass_prefix PASSED [100%]

============================== 3 passed in 0.31s ===============================

python -m ruff check src/transformers/cli/serving/utils.py tests/cli/test_serve.py
All checks passed!

git diff --check

Code Agent Policy

• I confirm that this is not a pure code agent PR.

Before submitting

• This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• Did you read the contributor guideline and the
  Pull Request checks?
• Was this discussed/approved via a Github issue or the forum? Please add a link
  to it if that's the case.
• Did you make sure to update the documentation with your changes according to the guidelines?
• Did you write any new necessary tests?

Who can review?

@Rocketknight1 @SunMarc

[github-actions]

---

CI recap

Dashboard: View test results in Grafana
Latest run: 28625420848:2
Result: success | Jobs: 2 | Tests: 7 | Failures: 0 | Duration: 25s