Add portable schema GUI, apply extensions, and deletion hardening

[christianbur]

Summary

Adds the portable schema GUI and apply workflow on top of the existing feature-branch backend: JSON import on the Custom Object Type add page, export tab and list export, choice_sets / objects apply support, and a reference demo document (security_objects.json).

Also includes operational hardening found while testing complex cross-COT schemas: COT deletion guards, bulk-delete reliability after dynamic model regeneration, and recursion guards for circular / polymorphic M2M graphs.

Portable schema import/export makes it straightforward for the community to share, reuse, and adopt Custom Object Type definitions as portable JSON documents.

Portable schema (GUI + apply)

• Import: JSON tab on Custom Object Types → Add (preview + apply via comparator/executor)
• Export: per-COT export tab + list action Portable schema download
• Apply extensions: document-level choice_sets and objects seed data (executor + validation)
• Documentation: portable-schema.md, rest-api.md, branch-aware apply (DDL in branch schema; COT/field metadata in main)
• Example: netbox_custom_objects/schema/examples/security_objects.json

Recommended top-level key order when authoring documents by hand:

schema_version → choice_sets → types → objects

Custom Object Type list & deletion

• List columns Objects and Referenced by (with bulk_load_list_stats)
• get_deletion_blockers() — block COT delete while instances exist or other COTs reference this type in schema (AbortRequest in UI/API)
• Safer COT delete when the dynamic backing table is already missing (drop tables before ContentType cleanup)

Bulk delete & dynamic model stability

Fixes ValueError: Must be "TableNModel" instance and related collector failures after cache_timestamp / model regeneration:

• Realign inbound/outbound M2M through FKs on all in-process class copies (registry, M2M descriptor, through cache)
• Polymorphic multiobject through source FK realignment (e.g. security-rb-demo1)
• CustomObject.delete() uses the canonical registered model class
• _netbox_private on CustomObject to avoid broken NetBox reverse-M2M changelog traversal
• Recursion guards during model generation (get_registered_model(), _model_generation_guard in get_models())

Other

• schedule_reindex_custom_object_type() — batch reindex enqueue on schema apply
• Tests: schema UI/API/executor, COT list stats, extended deletion scenarios (including security_objects apply smoke tests)

Test plan

• Apply security_objects.json via JSON tab (preview → apply)
• Export single COT + list portable schema export; verify slugs match
• COT list: Objects / Referenced by counts; delete blocked when > 0
• Bulk-delete security-service, security-service-group, and security-rb-demo1 instances after schema re-apply / cache invalidation
• COT bulk-delete with dependent types still referenced → blocking error (not silent schema break)
• python manage.py test netbox_custom_objects.tests.test_schema_ui netbox_custom_objects.tests.schema netbox_custom_objects.tests.test_cot_list_stats netbox_custom_objects.tests.test_deletion

Notes for reviewers

• Diff includes ~1k lines for security_objects.json (integration demo). Can trim or move to external reference if preferred.
• Some commits may include Co-authored-by: Cursor from local authoring; happy to squash on merge.
• Base branch: feature (not main).

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.

---

Christian seems not to be a GitHub user. You need a GitHub account to be able to sign the CLA. If you have already a GitHub account, please add the email address used for this commit to your account.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[jnovinger]

We require an approved issue before accepting PRs of this scope, and there isn't one linked here (or that I could find). This isn't a reflection on the quality of the work (the PR clearly represents significant effort), but taking on a large feature contribution without prior discussion means accepting maintenance responsibility for something we haven't had a chance to evaluate against our roadmap and architecture direction.

Please open an issue describing the feature and your proposed approach. Once we've had a chance to review and approve it, a PR will be welcome.