feat(model): add Groq chat model and formatter

[saswattulo]

feat(model): add groq chat model and formatter

AgentScope Version

1.0.19dev

Description

Background: Groq provides low-latency inference for popular open-weight
models (Llama, Mixtral, Gemma) via an OpenAI-compatible API. It was not
previously supported in AgentScope.

Changes made:

• src/agentscope/model/_groq_model.py — GroqChatModel class with
  streaming, tool use, and structured output support
• src/agentscope/formatter/_groq_formatter.py — GroqChatFormatter and
  GroqMultiAgentFormatter (inherits OpenAI-compatible formatter)
• pyproject.toml — added groq optional dependency
  (pip install agentscope[groq])
• tests/model_groq_test.py — 11 unit tests covering init, non-streaming,
  streaming, tool use, structured output, and usage parsing
• examples/agent/groq_agent/ — runnable example covering basic chat,
  streaming, ReAct agent with tools, and structured output

How to test:

pip install agentscope[groq]
export GROQ_API_KEY="your-key"
pytest tests/model_groq_test.py        # 11 tests, all pass
python examples/agent/groq_agent/main.py