Add agent identity to trace entries via persisted registry

[anormang1992]

Parent Issue

Sub-issue of #34 (Persist grounding traces to daily JSONL files)

Summary

Each trace entry should include a stable agent identifier so that traces from different agents are distinguishable. This requires a persisted agent registry that ensures the same agent instance receives the same ID across restarts and sessions.

Problem Statement

Issue #34 introduces trace persistence, but every trace entry is anonymous — there is no field identifying which agent produced it. In a single-agent setup this is tolerable, but it becomes a blocker for VRE Networks (multi-agent topologies sharing grounded knowledge across graphs).

If we ship trace persistence without agent identity:

• Historical traces will lack attribution, making them useless for multi-agent analysis
• Retrofitting agent IDs later means either backfilling or accepting a discontinuity in the trace history
• The VRE Workstation cannot filter or group activity by agent

Getting this right before traces are persisted at scale avoids a migration problem later.

Proposed Solution

Agent Identity Model

Introduce an AgentIdentity model (or similar) that holds:

• agent_id: a stable, unique identifier (e.g. UUID)
• name: a human-readable label (optional, for display purposes)
• created_at: when this agent was first registered

Persisted Registry

A lightweight registry that maps agent instances to their stable IDs:

• Storage: file-based (e.g. ~/.vre/agents.json or similar), consistent with Persist grounding traces to daily JSONL files #34's minimal-infrastructure approach
• Lookup: on VRE instantiation, the agent provides a registration key (e.g. name, config hash, or explicit ID) that resolves to a persisted identity
• Idempotent: subsequent instantiations with the same key return the same agent_id
• No graph mutation: the registry is external to the epistemic graph, like trace files

Trace Integration

Each trace entry (from #34) includes an agent_id field, populated from the registry at write time.

VRE Design Alignment

• Inspectability: Agent attribution makes traces more inspectable — you can answer "which agent attempted this grounding?" after the fact
• Minimal footprint: A single JSON file for the registry; no new infrastructure
• VRE Networks prerequisite: Multi-agent knowledge sharing requires agent identity as a foundational primitive; building it into the trace layer now avoids retrofit

Spike Notes

This will likely require spiking to resolve:

• What constitutes a stable "registration key" — agent name? Config fingerprint? Explicit user-supplied ID?
• Whether the registry should support deregistration or is append-only
• How identity interacts with VRE instantiation (constructor param? Separate registration call?)
• Whether identity should be optional (backward compat) or required when tracing is enabled

Acceptance Criteria

• AgentIdentity model exists with stable ID, optional name, and creation timestamp
• Persisted registry stores and retrieves agent identities idempotently
• VRE instance can be associated with an agent identity at construction or registration time
• Trace entries from Persist grounding traces to daily JSONL files #34 include the agent_id field when identity is configured
• Subsequent VRE instantiations with the same registration key resolve to the same agent_id
• Registry is file-based and requires no additional infrastructure
• Existing behavior is unaffected when no agent identity is configured

Dependencies

• Persist grounding traces to daily JSONL files #34 — trace persistence (this sub-issue adds a field to the trace schema)