fix: detect zombie browser context in getSession() to prevent 30s tab timeouts

[Love-JourneY]

Fix: proactive zombie context detection

Problem

getSession() at line 1142-1148 uses session.context.pages() as a health probe — but this is synchronous and returns cached data even when the underlying CDP connection is dead. When newPage() is subsequently called on a zombie context, it hangs for 30s.

Evidence

// browser pre-warmed successfully
{"ts":"2026-06-08T09:01:57.960Z","msg":"browser pre-warmed","ms":11505}

// 43 seconds later — first tab creation times out
{"ts":"2026-06-08T09:02:40.897Z","msg":"tab create failed","error":"tab create timed out after 30000ms"}

// Recovery — retry succeeds
{"ts":"2026-06-08T09:03:58.686Z","msg":"tab created","userId":"bbsoyarch"}

Root Cause

// server.js:1142-1148 — existing check
try {
    session.context.pages();  // sync — returns cached data even if CDP dead
} catch (err) {
    // only catches fully closed contexts
}
// ↓ newPage() hangs 30s on zombie context
const page = await session.context.newPage();  // line 2616

The project already tracks healthState.lastSuccessfulNav (line 633) and has an active health probe (lines 5949-5980) at 60s intervals. But between probes, a zombie context can go undetected.

Fix

Add a staleness check after the existing pages() probe — if no successful navigation in 120s, recreate the context proactively:

if (session && Date.now() - healthState.lastSuccessfulNav > 120_000) {
  log('warn', 'session context possibly stale, recreating', {
    userId: key,
    lastNavAgeMs: Date.now() - healthState.lastSuccessfulNav,
  });
  await closeSession(key, session, { reason: 'stale_context', ... });
  session = null;
}

Scope

• +11 / -0 lines, single file: server.js
• No new dependencies
• Leverages existing healthState.lastSuccessfulNav timestamp
• Backward compatible

Closes #5415