fix(orchestrator): open most recently active session, not newest-created

[AllBeingsFuture]

Problem

Clicking Orchestrator (sidebar / topbar / board / command palette) sometimes opened an empty conversation as if history had been wiped. History was still on another live orchestrator session; the UI/daemon had attached to a different, newer empty one.

Root cause

When more than one non-terminated orchestrator existed for a project (overlap during replacement, accidental duplicates, etc.), both the frontend picker (newestActiveOrchestrator) and the daemon ensure path (SpawnOrchestrator clean=false → newestSession) ranked by createdAt first.

That preferred a brand-new empty spawn over an older still-active orchestrator that held the real conversation — matching the “wrong session / empty history on open” report and the recency intent of #1362 (most recently used/active, not lex-smallest or newest-created).

Fix

Rank live orchestrators by:

1. activity.lastActivityAt (most recently used)
2. updatedAt
3. createdAt
4. session id

Applied in both:

• frontend/.../workspace.ts (newestActiveOrchestrator — all open entry points already use this)
• backend/.../service/session/service.go (sessionNewer — ensure/reuse when clean=false)

What was deliberately not changed

Path	Behavior preserved
SpawnOrchestrator(clean=false) with one active orchestrator	Still returns that session; no spawn
SpawnOrchestrator(clean=true) / restart / settings / restore-dialog	Still retires then spawns
Kill / terminate / multi-project isolation / worker spawn / review	Untouched
#3032 sidebar pinned Orchestrator row	Not duplicated; that remains a separate UX PR
#2501 false terminate + silent replace	Not fixed here (lifecycle); residual risk noted below

Tests

• Frontend: cd frontend && npm test -- --run src/renderer/types/workspace.test.ts — prefers recently-active over newer empty spawn
• Backend: cd backend && go test ./internal/service/session/ -count=1 — ensure returns most recently active among duplicates; existing clean=false/true tests still pass

Residual risks / follow-ups

• bug(lifecycle): orchestrator terminated mid-turn by a stale-activity/probe race, then silently replaced with no linkage or notification #2501: if the reaper falsely marks the live orchestrator terminated, ensure still spawns a cold session. Consecutive-probe work is in fix(lifecycle): require consecutive dead probes before terminating a session #2587; unlinked respawn notification is still open.
• Parent-orchestrator linkage (Orchestrator button on worker page links to wrong parent session #1211) is still absent in the Go domain; when multiple lives exist we pick by activity recency, not worker parent.

Closes nothing automatically; addresses the open-path half of the “click Orchestrator clears history” report and improves #1362 targeting.