feat: Add PromptBasedSGRAgent for models without structured output support

[Copilot]

SGRAgent uses OpenAI's response_format parameter for schema enforcement, but many models/architectures don't support this. Need alternative using prompt-based schema definition.

Changes

New Agent: PromptBasedSGRAgent

• Uses XML-tagged JSON in prompts instead of response_format parameter
• Parses <tool_call>{"name": "...", "arguments": {...}}</tool_call> from LLM responses
• Injects tool schemas directly into system prompt for better model understanding
• Maintains same execution flow as SGRAgent (reasoning → action → execution)

Prompt Template

• sgr_deep_research/core/prompts/prompt_based_system_prompt.txt with XML-style tool calling instructions
• Includes generated tool schemas with parameter types and descriptions

Core Methods

• _get_tools_schema(): Generates human-readable tool schemas from Pydantic models
• _parse_tool_call_from_response(): Regex extraction of JSON from XML tags with validation
• _prepare_context(): Dynamically builds system prompt with current tool availability

Usage

from sgr_deep_research.core.agents import PromptBasedSGRAgent

agent_def = AgentDefinition(
    name="prompt_based_sgr_agent",
    base_class=PromptBasedSGRAgent,
    # ... rest of config
)

Registered in AgentRegistry as "prompt_based_sgr_agent" for use in configuration files.

Testing

• 16 new tests covering initialization, schema generation, XML/JSON parsing, and integration
• All existing agent tests continue to pass

Original prompt

---

This section details on the original issue you should resolve

<issue_title>feat: prompt based schema definition</issue_title>
<issue_description>The default SGRAgent specifies output schema by setting response_format. But this is not supported for every model and/or architecture. Therefore we need an additional agent implementation that repeats SGRAgent behaviour but ensures output format by explicitly asking for in its prompt.

Excerpt from an example prompt:

...
You may call one or more functions to assist with the user query.
You are provided with function signatures within <tools></tools> XML tags:
<tools>
=> The same scheme for tools as we use in SGRAgent
</tools>

For each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:
<tool_call>
{"name": <function-name>, "arguments": <args-json-object>}
</tool_call>

</issue_description>

Comments on the Issue (you are @copilot in this section)

• Fixes feat: prompt based schema definition #1

---

💬 We'd love your input! Share your thoughts on Copilot coding agent in our 2 minute survey.