Smart relay backoff and refresh reliability fixes

[barrydeen]

Summary

• Connection attempt tracking: Each relay tracks connect() calls in a sliding 60s window. After 20 attempts, the relay enters a 5-minute cooldown. Successful connections reset the counter.
• Fix duplicate WebSocket connections on refresh: refreshFeed() was calling reconnectAll() right after updateRelays() which already connects new relays, causing duplicate WebSocket handshakes for relays mid-connect. Removed the redundant call and added a webSocket != null guard in connect().
• DM relays included in connected count: updateConnectedCount() was only counting persistent + ephemeral relays, missing DM relays entirely.
• Count refreshed after reconnectAll(): Explicitly recalculates after cleanup to avoid stale values.
• PullToRefreshBox: Native Material3 pull-to-refresh on the feed screen.
• Adaptive feed time window: Scales the since filter based on follow count (7 days for ≤20 follows down to 30 min for 700+).

Test plan

• Pull-to-refresh on feed — counter should accurately reflect all connected relays (persistent + DM + ephemeral)
• Simulate a failing relay (bad URL) — verify it enters 5 min cooldown after repeated failures, visible in relay console
• Manual reconnect from relay screen should clear backoff and reconnect immediately
• Small follow list (<20) should show older posts; large list (700+) should show recent only