refactor backend into modular packages and fix data pipeline bugs

Backend:
- Split monolithic database.go and handlers.go into focused modules
  (aggregate_queries, flow_queries, maintenance, schema, models,
   bandwidth_handlers, device_handlers, flow_handlers, stats_handlers, etc.)
- Add poll-time pre-aggregation (node pairs, bandwidth, traffic stats)
  with atomic commit via CommitPollResults transaction
- Add rolling in-memory cache for sub-second recent data queries
- Add device cache with periodic refresh from Tailscale API
- Add rate limiting middleware with per-IP token bucket
- Chunked backfill: split large time ranges into 30-min chunks to
  avoid API timeouts, with per-chunk progress commit
- Skip raw flow log insertion for historical data during backfill
  to prevent database bloat (was causing 2GB+ growth)
- Run cleanup before initial poll to purge stale data on startup
- Fix timestamp skew in log_converter: prefer "start" over "logged"
- Fix rolling cache UniquePairs: use max() instead of replace
- Fix nil bandwidth slice serializing as JSON null instead of []
- Add proper 405 handling for method-not-allowed routes
- Add schema tests and handler tests

Frontend:
- Smart header stats: fall back to stats store when network data
  unavailable (fixes zero stats on analytics page)
- BandwidthChart: abort stale fetch to prevent race condition
- FilterPanel: improved layout and responsiveness
- TimelineSlider: better touch/drag handling
- LogViewer/PortDetails: enhanced formatting and display
- Stats store: replace manual subscriptions with svelte get()
- Network store: refactored data flow and error handling
- Analytics page: improved chart layouts and interactions

Author: rajsinghtech

.gitignore

@@ -10,6 +10,7 @@ frontend/dist/
 backend/frontend/dist/
 backend/tsflow-backend
 backend/tsflow
+backend/tsflow-server*
 
 # Environment variables
 .env