Add DB retention/pruning to stop the database growing forever

[ashleigh-byte]

metrics/device_scans/speedtest rows accumulate indefinitely today -- existing queries only ever read a 24h window, but nothing ever deletes older rows, so a long-running install's DB grows without bound.

• New RETENTION_DAYS config (default 90 days)
• DB.prune_old_data() deletes metrics older than the retention window (cascading to their linked speedtest row via existing ON DELETE CASCADE), then removes any device_scans rows left orphaned by that cascade -- device_scans has no timestamp of its own, but every scan is created in the same transaction as exactly one metric via speedtest, so "no linked speedtest row remains" is an equivalent age check
• Runs once per detailed-report cycle (not every cycle -- no need to hit the DB with a cleanup pass every SLEEP_TIME tick)

Test plan:

• Unit-tested prune_old_data: old rows deleted and cascade correctly removes orphaned device_scans, recent rows survive untouched, idempotent re-run is a no-op, retention-day boundary (91 days deleted vs 89 days kept for a 90-day window)
• Verified get_metrics/get_metrics_with_device_counts still return correct results post-prune
• Unit-tested RETENTION_DAYS config validation (default, override, negative value rejected)