feat(hub): Model Registry client module

[abhijeet-dhumal]

Implement the hub client module wrapping kubeflow.hub.ModelRegistryClient (SDK 0.4.x). Completes the train -> register -> serve lifecycle within a single MCP session.

Context

• SDK 0.4.0 shipped ModelRegistryClient (v0.3.0+)
• Stub exists at kubeflow_mcp/hub/
• Without this, agents hit a dead end after "job complete"

Proposed Tools

Read-only:

• list_models, get_model, list_model_versions, get_model_version, compare_models

Mutating (confirm gate):

• register_model, promote_model, update_model_metadata

Acceptance Criteria

• Follows module contract (TOOLS, CLIENT_TOOL_DESCRIPTIONS, CLIENT_TOOL_ANNOTATIONS, INSTRUCTION_SECTIONS, CLIENT_RESOURCES, SECTION_ORDER)
• Confirm gate on mutations, persona filtering consistent with trainer
• Co-located tests, schema snapshot updated
• Resource: hub://guides/model-lifecycle

References

• SDK source: kubeflow/hub/api/model_registry_client.py
• ROADMAP Phase 6
• Related: Integrating the Kubeflow Spark Application with the Kubeflow SDK sdk#107

[abhijeet-dhumal]

/kind feature
/area hub

[myselfkunal]

/assign