SparkClient module for Kubeflow MCP Server

### Description

A `spark` client module for the Kubeflow MCP Server that enables AI agents to manage Spark Connect sessions and Spark batch jobs on Kubernetes through natural language, following the same modular architecture used by the `trainer` module.

This covers two complementary surfaces:
1. MCP Tools - wrapping `SparkClient` operations for session lifecycle, batch job management, and monitoring
2. Agent Skills - providing structured context (SKILL.md) that guides AI agents in helping users write, configure, debug, and deploy SparkClient code

### Use Case

_No response_

### Proposed Solution

_No response_

### Component

None