Update GkeCodeExecutor with option to use Agent Sandbox

[brandonroyal]

For users deploying ADK on Kubernetes and using GKE for Code Execution (https://google.github.io/adk-docs/tools/built-in-tools/#gke-code-executor), it would be great to leverage the faster startup latency and more granular control in Agent Sandbox (https://github.com/kubernetes-sigs/agent-sandbox)

I would like to see an additional type option in the GkeCodeExecutor which expands for the Sandbox type (https://github.com/kubernetes-sigs/agent-sandbox/blob/main/k8s/crds/agents.x-k8s.io_sandboxes.yaml). This would augment the current type: job which exists today. For backwards compatibility and stability, I'd expect the type = 'job' to be the default - at least until the Sandbox API graduates and stabilizes.

gke_executor = GkeCodeExecutor(
    ...
    type="sandbox"
    ...
)

When ADK instantiates a code execution request via the GkeCodeExecutor of type sandbox, ADK will create a SandboxClaim object. Once fulfilled with a Sandbox, ADK will execute the code and return the result. K8s administrators will have the option to configure things like WarmPools for low-latency execution and SandboxTemplates for ease of use. I will expect these administrator configuration will work as expected when SandboxClaims are instantiated via GkeCodeExecutor.

[aditya-shantanu]

@brandonroyal - I slightly alternate suggestion (in terms of implementation).
We're creating a Python SDK in AgentSandbox.
I'd like to see ADK directly use the Python SDK instead of us recoding very similar logic in two places.

[brandonroyal]

Using the general Python SDK as an intermediary makes sense.

[omarcevi]

@brandonroyal - I slightly alternate suggestion (in terms of implementation). We're creating a Python SDK in AgentSandbox. I'd like to see ADK directly use the Python SDK instead of us recoding very similar logic in two places.

Hey any updates regarding the python sdk. if possible i would like to work on this issue would this be ok? @seanzhou1023 @hangfei

[klateefa]

@brandonroyal Thanks for reporting!
This feature request is under review ,Thanks for your patience.