Add index health monitor workflow

[jpr5]

Summary

Adds a GitHub Actions cron workflow that probes both Pathfinder production instances every 4 hours for indexing health. Single new file, zero changes to application code.

• Liveness check via /health endpoint on mcp.copilotkit.ai and mcp.pathfinder.copilotkit.dev
• Per-source error detection for mapped sources (docs, code, ag-ui-docs, ag-ui-code, pathfinder-docs)
• Commit drift detection: compares indexed SHA against GitHub HEAD with 6h staleness threshold
• Chunk floor validation: 1000 for copilotkit-docs, 50 for pathfinder-docs
• Notification state machine: green/red transitions notify, green-to-green silent, red-to-red rate-limited 2x/24h
• State persistence via Actions cache (immutable save with run_id, prefix-match restore)
• Slack alerts via SLACK_WEBHOOK secret with instance details and run URL
• workflow_dispatch inputs for force_notify and dry_run

Why

We had no automated way to detect when Pathfinder indexes go stale, lose data (chunk count drop), or enter error state. This workflow closes that gap with minimal infrastructure — just a GitHub Action and a Slack webhook.

Test plan

• Trigger via workflow_dispatch with dry_run=true, verify probes run and no notifications sent
• Trigger via workflow_dispatch with force_notify=true, verify Slack message received
• Verify cron schedule fires every 4 hours
• Confirm SLACK_WEBHOOK secret is configured in repo settings