[Meta-driven Seed Loader] Engine must support dynamic relationship resolution & dependency ordering for seed data initialization

[hotlong]

Problem

ObjectStack currently allows objects and their fields to be defined and loaded entirely via metadata, but the engine's seed data loader does not automatically resolve relationships (lookup/master_detail fields) between records during initialization. Instead, it expects either pre-populated IDs, manual foreign-key mapping, or a fully correct insert order.

This makes onboarding, demo, and config scenarios fragile, and does not meet the self-driving data onboarding expectation of metadata-driven platforms.

Detail Checklist

• Engine seed loader must, during insert/upsert of seed data, automatically resolve lookup/master_detail references using target object's externalId (usually 'name'). This should happen if the field value is not already an internal ID, but is a string matching a unique key in the referenced object.
• Topological sort must be performed to determine import order based on object dependency graph, so parent objects are inserted before child objects that reference them.
• If circular or delayed references exist, loader must perform a second pass to fill unresolved relationships after all records are inserted.
• Engine should respect DatasetSchema's mode: upsert, not just blindly insert objects; upsert logic should be honored based on externalId for idempotency.
• Any reference resolution failure should generate actionable error messages with: field name, target object, externalId, attempted value, and possible missing referent.
• Loader should support dry-run for validation (check relationships before write) to surface broken references before failures.
• All logic should be metadata-driven and use SchemaRegistry, supporting hot-reload and dynamic object changes.

Acceptance Criteria

• Seed data can be written entirely using references via externalId (natural key).
• Demo and onboarding environments require no manual ID prep or table pre-sorting.
• All new objects with relationships work immediately from metadata alone, regardless of schema changes.
• Error handling and reporting matches the ease-of-use of platforms like Salesforce, Airtable, ServiceNow, etc.

Impact

This makes ObjectStack truly declarative, metadata-first for data onboarding, and closes a gap in platform intelligence compared to major low-code engines.