Correlating nested AI operations

When a new AI trace starts while another Observatory trace is active, the context manager preserves the parent snapshot.

The child trace stores propagated context metadata containing the parent trace and span IDs. When the child finishes, the parent trace context is restored so later events continue to correlate correctly.

This supports workflows such as:

Support agent
├── Model request
├── Nested classification agent
│   └── Model request
├── Tool call
└── Final model request

Nested agents remain separate traces rather than being rewritten into one database trace. The propagated metadata provides the cross-trace relationship.

Use AiObservatory::withContext() around the outer workflow to ensure user, tenant, feature, and custom tags are available to every nested operation.