refactor(integration): update Phase 2 tests to use MCP handlers

**Critical Fix**: Tests now properly test the MCP handler layer (the actual product) instead of raw API client.

**Changes**:
- All 15 tests now use `handleCreateWorkflow()` MCP handler
- Tests validate `McpToolResponse` structure (`success`, `data`, `error`)
- Created `mcp-context.ts` helper for configuring InstanceContext
- Fixed ERROR_HANDLING_WORKFLOW to add main connection (MCP validation requirement)
- Updated error/edge case tests to expect validation failures (correct MCP behavior)

**MCP Handler Validation**:
- Error scenarios now correctly expect `success: false` with validation errors
- Edge cases updated to reflect MCP handler's proper pre-validation
- Documents that MCP validation is CORRECT behavior (catches errors early)

**Test Results**: All 15 scenarios passing
- 8 valid workflow tests → expect `success: true`
- 7 validation tests (errors/edge cases) → expect `success: false`

**Why This Matters**:
AI assistants interact with MCP handlers, not raw API client. Testing the wrong layer would miss MCP-specific logic and validation.

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

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

tests/integration/n8n-api/utils/fixtures.ts

@@ -222,6 +222,7 @@ export const ERROR_HANDLING_WORKFLOW: Partial<Workflow> = {
       main: [[{ node: 'HTTP Request', type: 'main', index: 0 }]]
     },
     'HTTP Request': {
+      main: [[{ node: 'Handle Error', type: 'main', index: 0 }]],
       error: [[{ node: 'Handle Error', type: 'main', index: 0 }]]
     }
   },