03-integrations - [Add Claude Agent SDK agentic patterns: Orchestrator-Workers (subagents) and Hooks (tool governance)]

## Description

The `claude-agent/` directory currently has two examples:
- `claude-sdk/` — basic agent with 3 modes (basic, options, tools)
- `claude-with-code-interpreter/` — MCP tool integration with Code Interpreter

These cover basic usage but don't demonstrate key Claude Agent SDK features that other frameworks in the repo already showcase (e.g., OpenAI-agents has multi-agent handoff).

## Proposed additions

### 1. `claude-sub-agents/` — Orchestrator-Workers pattern

Demonstrates Claude Agent SDK's native subagent support (`AgentDefinition` + `Task` tool):
- Orchestrator agent delegates to specialized worker subagents
- `code-analyzer` subagent with read-only tools (Read, Grep, Glob)
- `reporter` subagent with write tools (Read, Write)
- Subagents run with isolated context and restricted tools

Maps to Anthropic's recommended [Orchestrator-Workers](https://www.anthropic.com/engineering/building-effective-agents) agentic pattern.

### 2. `claude-hooks/` — Tool governance with hooks

Demonstrates `PreToolUse` and `PostToolUse` hooks:
- `PreToolUse` blocks dangerous Bash commands and writes to system paths
- `PostToolUse` logs structured audit entries for every tool call
- README explains defense-in-depth: SDK hooks for built-in tools + AgentCore Policy for external tools behind Gateway

## Value

- Fills gaps vs other framework examples in the repo
- Showcases Claude Agent SDK features unique to the SDK (subagents, hooks)
- Connects to existing AgentCore Policy tutorials for a complete governance story
- Each example follows the same structure as existing examples (agent.py, Dockerfile, README, requirements.txt)

Start by comparing the existing claude-sdk/ and claude-with-code-interpreter/ examples, including their agent.py, Dockerfile, README, and requirements.txt structure. Then review the Claude Agent SDK documentation for AgentDefinition, Task, PreToolUse, and PostToolUse before adding claude-sub-agents/ and claude-hooks/. Done means both examples demonstrate the requested delegation, tool restrictions, blocking, and audit logging with clear README guidance.

Written by the indexing model from the issue text.