feat: add A2A multi-agent graph router research assistant sample

[johngiles12345]

Summary

• Adds a new use-case sample demonstrating the router-supervisor pattern using Strands Agents GraphBuilder with conditional edges, parallel dispatch, and AND-join semantics
• The router agent classifies requests and dynamically decides which sub-agents (research, analysis, writer) to invoke — supporting 4 distinct dispatch paths
• Includes both a local demo (single-process graph) and distributed deployment to Amazon Bedrock AgentCore using the A2A protocol

Architecture

Router → conditional dispatch → parallel workers → AND-join → synthesizer

• Router: Classifies request, outputs structured DISPATCH line (Claude Haiku 4.5)
• Research: Gathers information via search tools (Claude Haiku 4.5)
• Analysis: Analyzes data, compares sources (Claude Haiku 4.5)
• Writer/Synthesizer: Produces final report with AND-join guard (Claude Sonnet 4.6)

Key patterns demonstrated

• Conditional edge functions parsing structured router output
• AND-join semantics on Strands GraphBuilder (OR-native graph)
• A2A protocol sub-agents with SigV4-signed cross-runtime invocation
• Deployment via bedrock-agentcore-starter-toolkit Python SDK (no CDK/CloudFormation)

Test plan

• uv run python -m demo.run_demo --prompt "Research the impact of AI on healthcare and analyze the trends" runs successfully
• uv run python -m demo.run_demo --prompt "What are the latest developments in quantum computing?" (research-only path)
• uv run python -m demo.run_demo --prompt "Write a brief summary of cloud computing" (writer-only path)
• Deployment to AgentCore via uv run python -m deployment.deploy
• Cleanup via uv run python -m deployment.cleanup