AI workflow builder now prompts for real email addresses

When users asked n8n's AI builder to create workflows involving email or notifications, the system would silently hardcode fake addresses like user@example.com. The builder had a universal ban on placeholder() — a rule meant to prevent fake spreadsheet IDs, but it incorrectly blocked legitimate prompts for real user values. The result was workflows that looked complete but would silently fail at runtime.

This PR fixes the issue across four layers. First, the prompt rules now distinguish between discoverable resources where real IDs are required (spreadsheets, calendars) and user-provided values (email addresses, phone numbers) where placeholder() is the correct approach. The builder is explicitly told never to hardcode fake addresses.

Second, a new hasUnresolvedPlaceholders flag tracks workflows with incomplete user inputs and routes them to the setup wizard instead of running verification. Third, the setup service scans node parameters for placeholder sentinels and surfaces them as parameter issues so the wizard can collect real values. Fourth, the frontend wizard treats placeholder strings as unset values, prevents users from skipping them, and clears sentinels when real values are applied.

The fix introduces a reusable hasPlaceholderDeep() utility that recursively detects placeholder sentinels in strings, arrays, or objects. This runs everywhere placeholders need to be detected — in the workflow submission tool, the setup service, and the frontend setup wizard.

Workflows built by the AI now work on day one because users provide their actual email address, phone number, or notification target through the setup wizard instead of discovering the hardcoded fake value too late.