feat: add extra_resource_requests and extra_resource_limits to KubeflowExecutor

[ko3n1g]

Summary

• Adds extra_resource_requests and extra_resource_limits fields to KubeflowExecutor
• Allows users to specify arbitrary Kubernetes resource requests/limits beyond the existing GPU/CPU/memory fields (e.g. rdma/rdma_shared_device_a, hugepages-2Mi, custom device plugins)
• Both fields are dict[str, str] and are merged into the resource spec after the standard fields

Example

executor = KubeflowExecutor(
    ...
    extra_resource_requests={"rdma/rdma_shared_device_a": "1"},
    extra_resource_limits={"rdma/rdma_shared_device_a": "1"},
)

Test plan

• Existing tests pass: uv run -- pytest test/
• Manual e2e test on cluster with custom resource type

🤖 Generated with Claude Code