milestone(step): Complete Step 1 POC - Multi-agent memo orchestrator with web search and versioning

## Commit Body

### Overview
Implement complete multi-agent system for generating investment memos using LangGraph supervisor pattern with specialized agents. System achieved 7.5/10 quality score on real company test (Aalo Atomics) with web search enabled.

### Core Multi-Agent Architecture

**Agents Implemented:**
- Research Agent (basic): Claude-only research with structured output
- Research Agent (enhanced): Web search integration with Tavily/Perplexity
  - 4-query strategy: company overview, funding, team, news
  - Automatic synthesis of search results into structured JSON
- Writer Agent: Drafts 10-section memos following Hypernova template
- Validator Agent: Rigorous 0-10 scoring with specific feedback
- Supervisor: LangGraph orchestration with conditional routing

**State Management:**
- TypedDict schemas for type safety (MemoState, ResearchData, etc.)
- LangGraph StateGraph for workflow coordination
- Message passing between agents with append-only logs

### Web Search Integration

**Research Providers:**
- Tavily API: Recommended, 1000 free searches/month, AI summaries
- Perplexity API: Premium research with excellent citations
- Fallback: Claude-only mode (no API keys required)

**Multi-Query Strategy:**
Each company researched via 4 targeted searches:
1. Company overview (founders, technology, product)
2. Funding and investors (Crunchbase-style data)
3. Team backgrounds (LinkedIn-style data)
4. Recent news and announcements

Results synthesized by Claude into structured JSON matching memo requirements.

### Versioning System

**Semantic Versioning:**
- v0.0.x: Patch versions (automated iterations)
- v0.x.0: Minor versions (user-approved drafts)
- vx.0.0: Major versions (finalized memos)

**Version Tracking:**
- JSON-based version history (output/versions.json)
- Timestamps, scores, file paths for each iteration
- Auto-increment patch version on each run
- Manual promotion to minor/major (CLI command - TODO Week 4)

### Templates & Quality Standards

**Hypernova Template (10 Sections):**
1. Executive Summary
2. Business Overview
3. Market Context
4. Technology & Product
5. Traction & Milestones
6. Team
7. Funding & Terms
8. Risks & Mitigations
9. Investment Thesis
10. Recommendation

**Style Guide Enforcement:**
- Analytical tone (not promotional)
- Specific metrics over vague claims
- Source attribution requirements
- Balanced perspective (risks + opportunities)
- Good/bad examples for each section type

### CLI Interface

**Features:**
- Rich terminal output with colors and formatting
- Progress indicators during generation
- Validation score display with color coding
- Version history table (when >1 version exists)
- Markdown rendering of generated memos
- JSON state export for debugging

**Output Structure:**
- {company}-v0.0.1-draft.md: Memo content
- {company}-v0.0.1-state.json: Full workflow state
- versions.json: Cross-company version tracking

### Testing & Results

**Test Case: Aalo Atomics (Nuclear Energy Startup)**

Without Web Search (Baseline):
- Score: 3.5/10
- Issues: 90% '[DATA NOT AVAILABLE]' placeholders
- Result: Framework showing what should be evaluated

With Web Search (Enhanced):
- Score: 7.5/10
- Real data found:
  - Founders: Matt Loszak (CEO), Yasir Arafat (CTO, ex-INL)
  - Funding: $136M total across 3 rounds
  - Investors: Valor Equity Partners, NRG Energy, Hitachi
  - Technology: 50 MWe modular reactors for AI data centers
  - Milestones: DOE partnership, INL testing agreement
- Issues: Source citations need improvement, some promotional language
- Demonstrates 114% quality improvement with web search

### Project Structure

**Source Code:**
- src/agents/researcher.py: Basic Claude-only research
- src/agents/research_enhanced.py: Web search + synthesis
- src/agents/writer.py: Memo drafting
- src/agents/validator.py: Quality validation
- src/state.py: TypedDict schemas
- src/workflow.py: LangGraph orchestration
- src/versioning.py: Version tracking system
- src/main.py: CLI entry point

**Templates:**
- templates/memo-template.md: 10-section structure
- templates/style-guide.md: Hypernova writing standards

**Documentation:**
- README.md: Comprehensive project overview
- SETUP.md: Installation and usage guide
- docs/WEB_SEARCH_SETUP.md: Provider comparison
- changelog/2025-11-16_01.md: Detailed development log
- context-vigilance/: Design exploration documents

### Configuration

**Environment Variables:**
- ANTHROPIC_API_KEY (required)
- TAVILY_API_KEY or PERPLEXITY_API_KEY (optional, for web search)
- USE_WEB_SEARCH, RESEARCH_PROVIDER, MAX_SEARCH_RESULTS
- DEFAULT_MODEL (claude-sonnet-4-5-20250929)

### Dependencies Added

**Core:**
- langgraph>=0.2.0 (multi-agent orchestration)
- langchain>=0.3.0, langchain-anthropic>=0.2.0 (Claude integration)
- pydantic>=2.0.0 (data validation)

**Web Search:**
- tavily-python>=0.3.0 (web search)
- httpx>=0.27.0 (HTTP client)
- beautifulsoup4>=4.12.0 (HTML parsing)
- openai>=1.0.0 (for Perplexity API)

**CLI:**
- rich>=13.0.0 (terminal UI)
- python-dotenv>=1.0.0 (config management)

### Files Added (22 new files)

**Configuration:**
- .env.example
- .gitignore
- pyproject.toml
- SETUP.md

**Source Code (9 files):**
- src/__init__.py
- src/agents/__init__.py
- src/agents/research_enhanced.py
- src/agents/researcher.py
- src/agents/validator.py
- src/agents/writer.py
- src/main.py
- src/state.py
- src/versioning.py
- src/workflow.py

**Templates (2 files):**
- templates/memo-template.md
- templates/style-guide.md

**Documentation (7 files):**
- README.md (completely rewritten)
- docs/WEB_SEARCH_SETUP.md
- changelog/2025-11-16_01.md
- changelog/2025-11-16_02.md (this file)
- context-vigilance/Document-Examples.md
- context-vigilance/Generate-Investment-Memo-for-Portfolio-Company.md
- context-vigilance/Multi-Agent-Orchestration-for-Investment-Memo-Generation.md

**Test Data:**
- data/sample-company.json

### Known Limitations & Future Work

**Week 1 POC Limitations:**
- No revision loop (single-pass generation)
- No MCP servers (direct API calls only)
- Manual version promotion not implemented
- No checkpointing (cannot resume from failures)

**Next Steps:**

Week 2 (MCP Integration):
- Portfolio Data MCP server
- Crunchbase/PitchBook APIs via MCP
- Template serving via MCP
- Validation criteria as MCP resources

Week 3 (Revision Loop):
- Revision Agent implementation
- Iterative improvement to 8/10 threshold
- Specialized section writers
- Parallel section generation

Week 4 (Production):
- Web UI (Streamlit/Gradio)
- Human-in-the-loop checkpoints
- PDF export with branding
- Version comparison view
- Manual version promotion CLI

### Technical Decisions

**Why LangGraph?**
- Explicit state management (no hidden context)
- Conditional routing (validation score → finalize vs review)
- Built-in checkpointing support (Week 2)
- Visual debugging of agent transitions

**Why Tavily over Google/Bing?**
- Built specifically for AI agents
- Returns AI-generated summaries
- Better citation quality
- Generous free tier (1000/month)

**Why Semantic Versioning?**
- Clear iteration tracking (v0.0.x = drafts)
- User control over promotion (v0.x.0 = approved)
- Audit trail for all generations
- Enables A/B testing of different prompts

### Success Metrics

✅ Multi-agent orchestration functional
✅ Web search dramatically improves quality (+114%)
✅ Validator provides specific, actionable feedback
✅ Versioning enables iteration tracking
✅ Template/style guide enforced
✅ Complete documentation
✅ Tested on real company (Aalo Atomics)

cd /Users/mpstaton/code/lossless-monorepo/ai-labs/investment-memo-orchestrator

git add .
git status  # Verify what's being committed

git commit -F changelog/2025-11-16_02.md

# Or if you want to edit it first:
git commit -e -F changelog/2025-11-16_02.md

# Then push to remote
git push origin development  # or your branch name