feat(02): add MCP tool integration pattern with parallel fan-out pipeline

[ksopyla]

Summary

Adds Pattern 02: MCP Tool Integration -- exposes the crypto intelligence pipeline as an MCP tool with parallel fan-out/fan-in execution, alongside substantial infrastructure and documentation improvements.

Pattern 02 highlights

• 5-agent parallel pipeline: Research Planner → [News Scanner | Project Profiler | Community Analyst] → Intelligence Compiler
• Dual entry points: REST (POST /run on :8000) and MCP (research_crypto_project tool on :8001) running the same compiled LangGraph
• Data source ownership: each research node owns one external source (DuckDuckGo or CoinGecko), no duplication
• CoinGecko retry with exponential backoff, graceful degradation on partial failures
• Full test suite: unit, API, and e2e tests for the new pattern

Pattern 01 improvements

• Renamed 01-multi-agent-single-system → 01-orchestrator-pipeline to match curriculum terminology
• Added proper test suite (unit, API, e2e) and endpoints.http

Infrastructure & tooling

• Shared base Docker image (infra/docker/base/Dockerfile.agent) for all examples
• Per-directory mypy wrapper (scripts/linting/run_mypy.py) to avoid duplicate module errors in the monorepo
• Test suite runner (scripts/testing/run_test_suite.py) with auto-discovery
• Updated CI workflows to use the wrapper scripts
• New .cursor/mcp.json, skills, and rules for development workflow

Documentation

• New docs/vision.md and expanded docs/curriculum.md with full pattern progression
• CHANGELOG.md and AGENTS.md with architecture decisions
• Comprehensive README for both patterns with Mermaid diagrams
• Research reports on agent server alternatives and distributed agent engineering

Shared library (libs/common)

• Standardized LangSmith tracing workflow with configure_tracing()
• Config and LLM factory improvements

Test plan

• uv run python scripts/testing/run_test_suite.py passes for all examples
• uv run python scripts/linting/run_mypy.py passes
• uv run ruff check . and uv run ruff format --check . pass
• docker compose up --build works for both examples 01 and 02
• MCP endpoint at :8001/sse is discoverable by Claude Desktop / Cursor
• REST endpoint POST /run returns full intelligence report with intermediate artifacts