refactor(tools): wrap sync code interpreter tools with asyncio.to_thread

[zhangzx-uiuc]

Supporting Async CodeInterpreterToolkit

This PR updates the existing synchronous CodeInterpreterToolkit to an async implementation.

At the moment, aioboto3 does not support BedrockAgentCore. As a temporary solution, we wrap the synchronous Code Interpreter and terminal execution logic using asyncio.to_thread, exposing an async interface while preserving existing behavior. This allows the toolkit to integrate cleanly with async agent pipelines without blocking the event loop.

---

Testing

We tested the async implementation on AIME-2024 using Qwen3-8B, hosted on a single H200 node.

Command used:

strands-env eval aime-2024 \
    --env examples/eval/aime_code/code_sandbox_env.py \
    --backend sglang \
    --base-url http://localhost:30000 \
    --n-samples-per-prompt 1 \
    --temperature 0.0 \
    --max-concurrency 30

Results

Version	Running Time	Performance
Async	193s	53.3%
Sync	205s	53.3%

Summary

• The async implementation is slightly faster than the synchronous version under this setting.
• While the AIME evaluation set is relatively small, the async design is expected to yield larger efficiency gains in large-scale coding agent training and higher-concurrency scenarios.

[zhangzx-uiuc]

Hi team, please review this PR to enable async code interpreter toolkit. Feel free to leave any comments and suggestions. Thanks!