Phase 3: AI pipeline generation and interactive wizard

[netsirius]

Summary

Adds AI-powered pipeline generation from natural language and an interactive step-by-step wizard.

AI Generation

• LLMClient: HTTP client for Claude and OpenAI APIs using java.net.http (zero external dependencies)
• PromptBuilder: schema-grounded prompts with full JSON Schema + connector reference
• GenerateCommand: weaver generate "<description>" with validation loop (generate → validate → retry, max 3 attempts)
• WeaverConfig: ~/.weaver/config.yaml for AI provider, model, API key configuration

Interactive Wizard

• InteractiveWizard: step-by-step pipeline creation without LLM (weaver init --interactive)
• Asks source type, connection, query, transforms, quality checks, sink type
• Generates valid YAML with inline tests

Project Scaffolding

• InitCommand: weaver init <project-name> creates project structure with example pipeline, connections.yaml, .env.example, .gitignore

Pipeline JSON Schema

• core/src/main/resources/schemas/pipeline.schema.json for IDE autocomplete and LLM prompt grounding

New CLI commands

weaver init my-project              # Scaffold new project
weaver init --interactive            # Step-by-step wizard
weaver generate "description"        # AI pipeline generation

Test plan

• sbt compile — All modules compile
• sbt core/test — 47 tests pass
• Manual: weaver init test-project creates project structure
• Manual: weaver generate "read CSV, filter, write parquet" with ANTHROPIC_API_KEY set