Agent publish readiness validation

Prevent an agent from being published when it cannot run successfully. Before publishing, validate:

• The selected AI model exists and supports the required capabilities.
• At least one complete operation is enabled.
• Configured tools and referenced resources still exist.
• Escalation rules have valid reviewers.
• Knowledge sources are ready.
• Prompt tool references are correctly configured. Show actionable errors and block publication until P0 problems are resolved. Versioning and rollback should be tracked separately.