Add ZMQ hashblock subscriber for sub-second chain refresh

New internal/zmqmon package subscribes to bitcoind's hashblock
ZMQ topic and emits TipEvents. Uses github.com/go-zeromq/zmq4
(pure Go, builds with CGO_ENABLED=0). Exponential backoff on
connection failure so bitcoind restarts don't kill the
subscriber permanently; event channel drops rather than blocks
if the consumer is slow (signals are advisory, not logs).

Aggregator.Run now takes a <-chan TipEvent; when a tip arrives
it fires an immediate refresh() outside the normal ticker
cadence. With a 5s poll interval and 0.5-1s ZMQ latency from
bitcoind, dashboards now reflect new tips roughly 4x faster.

Endpoint comes from BITCOIN_ZMQ_BLOCK — empty disables ZMQ
entirely and the aggregator just runs on the ticker alone.

api/cmd/kamado-api/main.go

@@ -22,6 +22,7 @@ import (
 	"github.com/kamadopool/kamado-api/internal/logmon"
 	"github.com/kamadopool/kamado-api/internal/state"
 	"github.com/kamadopool/kamado-api/internal/store"
+	"github.com/kamadopool/kamado-api/internal/zmqmon"
 )
 
 func main() {
@@ -69,9 +70,14 @@ func main() {
 	// real-time updates without polling.
 	agg.OnRefresh = api.Hub.Broadcast
 
-	go agg.Run(ctx)
+	// Optional: bitcoind hashblock ZMQ subscription for sub-second
+	// chain-tip refreshes. Empty endpoint disables it.
+	zmq := zmqmon.New(cfg.BitcoinZMQBlock, log)
+	go zmq.Run(ctx)
 
-	// Tail the ckpool log for block-solve events.
+	go agg.Run(ctx, zmq.Events)
+
+	// Tail the ckpool log for block-solve events (our own solves).
 	tailer := logmon.New(cfg.CKPoolLogFile, log)
 	go tailer.Run(ctx)
 	go agg.IngestBlockEvents(ctx, tailer.Events)