fix: comprehensive bug fixes and architecture restructuring

Phase 1 — Bug fixes (7 bugs):
- Bug 1: Pulsar mode data loss — always persist to SQLite regardless of mode
- Bug 2: PulsarSend closing TCP client — use independent resetPulsarProducer()
- Bug 3: Greedy regex — use non-greedy (?s)ZCZC.*?NNNN
- Bug 4: Data after NNNN discarded — keep remaining buffer data
- Bug 5: strings.Index > 0 boundary — use strings.Contains
- Bug 6: Variable shadowing in PulsarSend — use = not :=
- Bug 7: Accept failure nil panic — add continue + retry logic

Phase 2 — Architecture restructuring:
- Split utils/ into config/, serial/, telegram/, storage/, transport/
- Define Sender, Repository, Reader interfaces
- Introduce app/ layer with context.Context lifecycle
- Replace spinlock with sync.Mutex
- Unified Config struct replaces 15+ global vars

Phase 3 — Testing & tooling:
- telegram/parser_test.go (7 test cases)
- storage/store_test.go (5 test cases)
- transport/transport_test.go (4 test cases)
- Taskfile: add test, test-race, test-cover tasks
- Go version: 1.15 -> 1.21

utils/socket.go

@@ -23,16 +23,29 @@ func Listen(address string) {
 	}
 	defer server.Close()
 	Log.Info("listen on ", zap.String("address", address))
+
+	acceptFailCount := 0
+	const maxAcceptFail = 3
+
 	for ServerRunning {
 		conn, err := server.Accept()
 		if err != nil {
-			ServerRunning = false
-			Log.Error("error in create connection ", zap.Error(err))
+			acceptFailCount++
+			Log.Error("error in create connection ",
+				zap.Error(err),
+				zap.Int("failCount", acceptFailCount))
+			if acceptFailCount >= maxAcceptFail {
+				ServerRunning = false
+				Log.Fatal("accept failed too many times, shutting down")
+			}
+			continue
 		}
+		acceptFailCount = 0
+
 		if client == nil {
 			client = conn
 			ClientReady = false
-			Log.Info("client connected on ", zap.Any("address", client.RemoteAddr()))
+			Log.Info("client connected on ", zap.Any("address", conn.RemoteAddr()))
 			LoadUnprocessed()
 		} else {
 			Log.Info("remove old client")