feat(tasks): phase 3 — close the knowledge loop (complete_task + retrieval)

Adds the compounding knowledge loop the task model is built around:

- complete_task(outcome, summary): a nomos-LOCAL, session-scoped tool (the
  shared MCP server has no session id). Introduces the local-tool mechanism —
  buildTools appends task tools, the agent loop routes them to handleTaskTool
  instead of the MCP client. Sets the task's terminal status/outcome/summary,
  mirrors it onto the task entity, and emits task.status.
- Knowledge → task linkage: after a successful upsert_knowledge in a task,
  nomos links the note to the task entity (documents) and emits
  knowledge.recorded, so the task's outcome view shows what it learned. The
  note's about-link to the involved entity (written by upsert_knowledge) is the
  retrieval path future tasks use.
- SOUL: every chat is a task loop — retrieve prior knowledge FIRST
  (get_entity_knowledge on the target), plan, execute, record learnings, then
  complete_task. Scales down for trivial read-only tasks.
- deleteSession now cleans up the task entity, its relationships, and its
  task-scoped events (was orphaning them); the knowledge doc itself and its
  about-links survive, as knowledge should outlive the task.

Verified end-to-end: a task recorded a note and completed; task.status +
knowledge.recorded hit the SSE stream; status=done/outcome=success persisted;
the note linked to both lxc:caddy (retrieval) and the task; a future
get_entity_knowledge(lxc:caddy) surfaces it; delete cleaned edges+events (0/0/0)
while the knowledge survived.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

cmd/nomos/agent.go

@@ -360,7 +360,15 @@ func (a *agent) chatWith(ctx context.Context, sessionID, message, systemInject s
 			})
 
 			start := time.Now()
-			result, callErr := a.client.callTool(tc.Function.Name, args)
+			// Session-scoped task tools are handled in-process; everything else
+			// is forwarded to the shared MCP server.
+			var result any
+			var callErr error
+			if localRes, handled := a.handleTaskTool(ctx, sessionID, tc.Function.Name, args); handled {
+				result = localRes
+			} else {
+				result, callErr = a.client.callTool(tc.Function.Name, args)
+			}
 			elapsed := int(time.Since(start).Milliseconds())
 
 			inputJSON, _ := json.Marshal(args)
@@ -397,6 +405,12 @@ func (a *agent) chatWith(ctx context.Context, sessionID, message, systemInject s
 			// results — so a bulk query doesn't drag the whole fleet in.
 			a.store.recordTouched(ctx, sessionID, tc.Function.Name, args)
 
+			// When the agent records knowledge, link that note to this task so
+			// the task's outcome view shows what it learned (and pulse it live).
+			if tc.Function.Name == "upsert_knowledge" {
+				a.store.linkKnowledgeToTask(ctx, sessionID, string(resultJSON))
+			}
+
 			emit(agentEvent{
 				Type:      "tool_result",
 				Data:      map[string]any{"name": tc.Function.Name, "result": result, "id": tc.ID},
@@ -619,6 +633,9 @@ func (a *agent) buildTools() ([]openai.ChatCompletionToolParam, error) {
 	if err != nil {
 		return nil, err
 	}
+	// Append nomos-local, session-scoped task tools (complete_task, …) to the
+	// MCP tool list. They're routed to handleTaskTool, not the MCP client.
+	defs = append(defs, taskToolDefs()...)
 
 	var tools []openai.ChatCompletionToolParam
 	for _, d := range defs {