Phase 7: Agent Selector — per-agent soul.md + ChromaDB memory filtering

- Agent dropdown UI (chip button + hidden select) in composer header
- Session.agent field persists agent selection across refresh
- soul.md loaded per-agent via ephemeral_system_prompt injection
- ChromaDB memory filtered by agent topic (lotus/, sunflower/, etc.)
- Fixed streaming.py: agent→_ai_agent variable shadowing (lines 1161, 1163)
- New API endpoints: /api/agents/topology, /api/agents/memory/search
- Agent metadata registry with emoji, name, description per Tier-2 agent

api/models.py

@@ -45,6 +45,7 @@ class Session:
                  input_tokens: int=0, output_tokens: int=0, estimated_cost=None,
                  personality=None,
                  active_stream_id: str=None,
+                 agent: str=None,
                  pending_user_message: str=None,
                  pending_attachments=None,
                  pending_started_at=None,
@@ -68,6 +69,7 @@ class Session:
         self.estimated_cost = estimated_cost
         self.personality = personality
         self.active_stream_id = active_stream_id
+        self.agent = agent
         self.pending_user_message = pending_user_message
         self.pending_attachments = pending_attachments or []
         self.pending_started_at = pending_started_at
@@ -103,6 +105,7 @@ class Session:
             'title': self.title,
             'workspace': self.workspace,
             'model': self.model,
+            'agent': getattr(self, 'agent', None),
             'message_count': len(self.messages),
             'created_at': self.created_at,
             'updated_at': self.updated_at,