Polar integration

[HaoZhang534]

Summary

This PR adds a Polar Responses API agent adapter for Gym under responses_api_agents/polar.

The adapter converts Gym /run requests into the single-row SWE-Gym JSONL format consumed by ProRL-Agent-Server/examples/swegym_slime_grpo/simple_inference.py, executes Polar inference, and converts the result JSON back into a Gym NeMoGymResponse.

Main additions:

• Add Polar agent server implementation with /responses and /run support.
• Support both direct inference against an existing OpenAI-compatible SGLang server and SLURM-backed inference through the Polar submit script.
• Preserve tool calls, tool outputs, text outputs, token ids, logprobs, usage, and Polar metadata in the returned response.
• Add a Polar Gym config, example SWE-Gym row, README usage docs, and unit tests.

Testing

.venv/bin/pytest responses_api_agents/polar/tests/test_app.py

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.