refactor: replace tiered aggregation with flat storage

[rajsinghtech]

Summary

• Replaces 13-table tiered scheme (*_minutely, *_hourly, *_daily, flow_logs_current) with 5 flat tables: node_pairs, bandwidth, bandwidth_by_node, traffic_stats, poll_state
• Adds idempotent migration in Init() — renames existing minutely tables and drops old tiers, preserving data on upgrade
• Query-time bucketing in GetBandwidth, GetNodeBandwidth, GetTrafficStats via resolveBucketSize(): ≤2h→1min, ≤48h→1hour, else→1day — no information loss from pre-aggregation
• Removes raw flow log storage (flow_logs_current + InsertFlowLogs); GET /api/flow-logs returns empty
• Single TSFLOW_RETENTION env var (default 720h / 30 days) replaces TSFLOW_RETENTION_MINUTELY, TSFLOW_RETENTION_HOURLY, TSFLOW_RETENTION_DAILY

Test Plan

• go test ./internal/... passes (database, handlers, services)
• Fresh install: verify Init() creates 5 flat tables, no old tables exist
• Upgrade: run against a DB with old *_minutely tables — verify data is preserved in renamed tables and old tiers are dropped
• Analytics page loads and shows correct data across all time ranges (1h, 6h, 24h, 7d, 30d)
• Bandwidth chart buckets correctly for wide time ranges
• TSFLOW_RETENTION=168h env var respected at startup