[serve] Forward tool_calls/tool_call_id in processor inputs

[qgallouedec]

• get_processor_inputs_from_messages builds a new dict per message with only role and content, dropping tool_calls and tool_call_id. This means apply_chat_template can't properly render multi-turn tool-use conversations.
• Forward these fields so the chat template sees the full tool-use context.

Example

Multi-turn tool-use conversation sent to transformers serve:

from transformers.cli.serving.utils import BaseHandler, Modality

messages = [
    {"role": "user", "content": "What's the weather in Paris?"},
    {"role": "assistant", "tool_calls": [{"id": "call_1", "type": "function", "function": {"name": "get_weather", "arguments": '{"city": "Paris"}'}}]},
    {"role": "tool", "content": "22°C, sunny", "tool_call_id": "call_1"},
]
print(BaseHandler.get_processor_inputs_from_messages(messages, Modality.LLM))

Before this PR, tool_calls and tool_call_id were dropped:

[
    {'role': 'user', 'content': "What's the weather in Paris?"},
    {'role': 'assistant', 'content': []},
    {'role': 'tool', 'content': '22°C, sunny'}
]

After this PR, they are forwarded to apply_chat_template:

[
    {'role': 'user', 'content': "What's the weather in Paris?"},
    {'role': 'assistant', 'content': [], 'tool_calls': [{'id': 'call_1', 'type': 'function', 'function': {'name': 'get_weather', 'arguments': '{"city": "Paris"}'}}]},
    {'role': 'tool', 'content': '22°C, sunny', 'tool_call_id': 'call_1'}
]

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.