Optimize relationship visibility and org path handling

[yaojin3616]

Summary

• support multi-select human and agent relationship management with searchable candidate lists
• centralize agent visibility rules and restrict visibility to same-tenant accessible agents only
• rebuild org department/member paths from department hierarchy and add backfill support
• redirect tenant switching to the homepage instead of preserving the current route

Validation

• PYTHONPATH=backend pytest backend/tests/test_agent_visibility.py -q
• PYTHONPATH=backend pytest backend/tests/test_org_sync_adapter.py -q
• python3 -m py_compile backend/app/core/permissions.py backend/app/api/agents.py backend/app/api/relationships.py backend/app/services/org_sync_adapter.py backend/app/api/enterprise.py backend/app/scripts/backfill_department_paths.py
• npm run build

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 3b5f8706ad

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[chatgpt-codex-connector]

Normalize target IDs before filtering self relationships

The new self-reference guard compares item.target_agent_id to str(agent_id) as raw strings, but save_agent_relationships no longer re-checks after parsing to UUID. If a client sends an equivalent UUID in a different string form (e.g., uppercase or brace-wrapped), this check is bypassed and the API can persist an agent-to-itself relationship, which the previous implementation explicitly prevented.

Useful? React with 👍 / 👎.

[chatgpt-codex-connector]

Keep parent linkage resilient to unordered department sync

This change drops the provider-side parent hint from path and relies on post-sync rebuilding from parent_id, but parent_id is only set when the parent row already exists during upsert. If departments arrive child-before-parent (the WeCom adapter consumes API order directly), children are saved with parent_id=None, and the rebuild then writes flattened top-level paths, which propagates incorrect org paths to members and breaks hierarchical department filtering until a later sync happens to reorder correctly.

Useful? React with 👍 / 👎.