context management API by code-interpreter

[Pangjiping]

Why do you need it?

Through systematic integration testing, we discovered that during concurrent runcode operations in code-interpreter, there are occasional issues where Jupyter becomes unresponsive.

We suspect that too many contexts were generated during testing, causing the Jupyter server to stop responding. Additionally, to standardize user behavior, we also want to have a mechanism to expire/clean up contexts.

How could it be?

• a new API DELETE /context/<context-id> by code-interpreter which could delete a existed context thread.
• a new API DELETE /context/language/<language> by code-interpreter which could delete all existed contexts under target language.
• a new API GET /contexts?language=<language> by code-interpreter which could list all active/available contexts
• SDK update
• API spec yaml update

Other related information

• https://github.com/alibaba/OpenSandbox/actions/runs/20460014266/job/58790425298

[ninan-nn]

LGTM. One suggestion on the path naming to keep it consistent: let's use the plural /contexts for all endpoints, especially for DELETE:

Individual: /context/{id} -> /contexts/{id}

By Language: /context/language/{lang} -> /contexts?language={lang} (This aligns better with your GET /contexts design).

[Pangjiping]

• TEST: integration test for context management #83 integration test for SDK
• feat(components/execd): add new api for context management #48 context management in execd

[liuxiaopai-ai]

Added a focused follow-up fix in PR #373: #373

The broader context-management API from this issue already exists in the codebase, but missing context lookups could still fall through an unsafe path. This PR makes execd return 404 / CONTEXT_NOT_FOUND for missing contexts and adds runtime/controller tests for that behavior.