[recipe] An agent-lightning like RL training pipeline

[linxxx3]

What does this PR do?

Stage 1 PR for RFC #3434

Checklist Before Starting

• Search for similar PRs. Paste at least one query link here: ...
• Format the PR title as [{modules}] {type}: {description} (This will be checked by the CI)
  • {modules} include fsdp, megatron, sglang, vllm, rollout, trainer, ci, training_utils, recipe, hardware, deployment, ray, worker, single_controller, misc, perf, model, algo, env, tool, ckpt, doc, data
  • If this PR involves multiple modules, separate them with , like [megatron, fsdp, doc]
  • {type} is in feat, fix, refactor, chore, test
  • If this PR breaks any API (CLI arguments, config, function signature, etc.), add [BREAKING] to the beginning of the title.
  • Example: [BREAKING][fsdp, megatron] feat: dynamic batching

Test

There are some unit tests in recipe/agent_lightning_like/test.

Also a gsm8k tool agent example is here for e2e test: recipe/agent_lightning_like/example/run_qwen2.5_7b.sh

local training curves:

API and Usage Example

see recipe/agent_lightning_like/example/README.md

Design & Code Changes

see RFC #3434

Checklist Before Submitting

Important

Please check all the following items before requesting a review, otherwise the reviewer might deprioritize this PR for review.

• Read the Contribute Guide.
• Apply pre-commit checks: pre-commit install && pre-commit run --all-files --show-diff-on-failure --color=always
• Add / Update the documentation.
• Add unit or end-to-end test(s) to the CI workflow to cover all the code. If not feasible, explain why: ...
• Once your PR is ready for CI, send a message in the ci-request channel in the verl Slack workspace. (If not accessible, please try the Feishu group (飞书群).)

[gemini-code-assist]

Code Review

This pull request introduces a new recipe for an agent-lightning like RL training pipeline. The changes are extensive, adding a complete workflow from data handling, agent interaction, to training. My review focuses on correctness and robustness. I've identified several critical issues, including potential race conditions, resource leaks, and runtime errors due to incorrect exception handling and unsafe coding patterns. I've also pointed out a significant performance issue in the example agent client. Addressing these points will greatly improve the stability and reliability of the new training pipeline.