feat: implement Phase 0 TDD autopilot dry-run foundation

Implements the complete Phase 0 spike for autonomous TDD workflow with orchestration architecture.

## What's New

### Core Services (tm-core)
- **PreflightChecker**: Validates environment prerequisites
  - Test command detection from package.json
  - Git working tree status validation
  - Required tools availability (git, gh, node, npm)
  - Default branch detection

- **TaskLoaderService**: Comprehensive task validation
  - Task existence and structure validation
  - Subtask dependency analysis with circular detection
  - Execution order calculation via topological sort
  - Helpful expansion suggestions for unready tasks

### CLI Command
- **autopilot command**: `tm autopilot <taskId> --dry-run`
  - Displays complete execution plan without executing
  - Shows preflight check results
  - Lists subtasks in dependency order
  - Preview RED/GREEN/COMMIT phases per subtask
  - Registered in command registry

### Architecture Documentation
- **Phase 0 completion**: Marked tdd-workflow-phase-0-spike.md as complete
- **Orchestration model**: Added execution model section to main workflow doc
  - Clarifies orchestrator guides AI sessions vs direct execution
  - WorkflowOrchestrator API design (getNextWorkUnit, completeWorkUnit)
  - State machine approach for phase transitions

- **Phase 1 roadmap**: New tdd-workflow-phase-1-orchestrator.md
  - Detailed state machine specifications
  - MCP integration plan with new tool definitions
  - Implementation checklist with 6 clear steps
  - Example usage flows

## Technical Details

**Preflight Checks**:
- ✅ Test command detection
- ✅ Git working tree status
- ✅ Required tools validation
- ✅ Default branch detection

**Task Validation**:
- ✅ Task existence check
- ✅ Status validation (no completed/cancelled tasks)
- ✅ Subtask presence validation
- ✅ Dependency resolution with circular detection
- ✅ Execution order calculation

**Architecture Decision**:
Adopted orchestration model where WorkflowOrchestrator maintains state and generates work units, while Claude Code (via MCP) executes the actual work. This provides:
- Clean separation of concerns
- Human-in-the-loop capability
- Simpler implementation (no AI integration in orchestrator)
- Flexible executor support

## Out of Scope (Phase 0)
- Actual test generation
- Actual code implementation
- Git operations (commits, branches, PR)
- Test execution
→ All deferred to Phase 1

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>

.taskmaster/docs/tdd-workflow-phase-0-spike.md

@@ -1,8 +1,13 @@
-# Phase 0: Spike - Autonomous TDD Workflow
+# Phase 0: Spike - Autonomous TDD Workflow ✅ COMPLETE
 
 ## Objective
 Validate feasibility and build foundational understanding before full implementation.
 
+## Status
+**COMPLETED** - All deliverables implemented and validated.
+
+See `apps/cli/src/commands/autopilot.command.ts` for implementation.
+
 ## Scope
 - Implement CLI skeleton `tm autopilot` with dry-run mode
 - Show planned steps from a real task with subtasks