Make Chrome tab/session handling more robust (leaks, crashes, concurrency edge cases)

[aplgr]

Goal

Improve renderer stability under concurrency: prevent leaks, handle crashes, and avoid “stuck” Chrome sessions.

Why

Headless Chrome is powerful but brittle at scale. Most production issues will be lifecycle + concurrency related.

Scope

• Ensure every request reliably:
  • creates/borrows a page/session
  • closes it on success and on every error path
  • respects context cancellation
• Add safeguards:
  • max concurrent pages
  • watchdog: restart browser on repeated failures
  • detect and recover from “Chrome died” mid-request
• Add lightweight instrumentation:
  • counters for active pages
  • render duration histogram (if metrics exist)

Acceptance criteria

• No steady growth in goroutines / memory under sustained load test
• Renderer survives Chrome crash (auto-recover within a bounded time)
• Concurrency limits are enforced (fails fast with 503/429 instead of melting)

Notes / hints

Pair this with the improved load test issue so you can validate fixes.