Improve logging and error handling, (#5)

* Improve logging and error handling

This overhauls the logging and error responses from the *Conn. Namely, it
changes the logging backend to `log/slog` which makes it easier to manage the
library's logging centrally (however the old functionally to provide a custom
logger is retained). It also captures the reason why a specific connection to a
ZK server was closed in the error that is propagated to all callers. This can be
checked with `errors.Is`.

This also adds an exponential backoff feature to the client, which will make it
back off when reconnecting to a ZK server.

* Remove logger option, it messes with the logging

* Change default behavior of DNSHostProvider

* Split implementation of HostProvider, fix up tests

cluster_test.go

@@ -1,6 +1,7 @@
 package zk
 
 import (
+	"log/slog"
 	"sync"
 	"testing"
 	"time"
@@ -112,10 +113,10 @@ func TestNoQuorum(t *testing.T) {
 		t.Fatalf("Failed to connect and get session")
 	}
 	initialSessionID := zk.sessionID
-	DefaultLogger.Printf("    Session established: id=%d, timeout=%d", zk.sessionID, zk.sessionTimeoutMs)
+	t.Logf("Session established: id=%d, timeout=%d", zk.sessionID, zk.sessionTimeout)
 
 	// Kill the ZooKeeper leader and wait for the session to reconnect.
-	DefaultLogger.Printf("    Kill the leader")
+	t.Logf("Kill the leader")
 	disconnectWatcher1 := sl.NewWatcher(sessionStateMatcher(StateDisconnected))
 	hasSessionWatcher2 := sl.NewWatcher(sessionStateMatcher(StateHasSession))
 	tc.StopServer(hasSessionEvent1.Server)
@@ -135,7 +136,7 @@ func TestNoQuorum(t *testing.T) {
 	}
 
 	// Kill the ZooKeeper leader leaving the cluster without quorum.
-	DefaultLogger.Printf("    Kill the leader")
+	t.Logf("Kill the leader")
 	disconnectWatcher2 := sl.NewWatcher(sessionStateMatcher(StateDisconnected))
 	tc.StopServer(hasSessionEvent2.Server)
 
@@ -151,7 +152,7 @@ func TestNoQuorum(t *testing.T) {
 	// Make sure that we keep retrying connecting to the only remaining
 	// ZooKeeper server, but the attempts are being dropped because there is
 	// no quorum.
-	DefaultLogger.Printf("    Retrying no luck...")
+	t.Logf("Retrying no luck...")
 	var firstDisconnect *Event
 	begin := time.Now()
 	for time.Now().Sub(begin) < 6*time.Second {
@@ -269,7 +270,7 @@ func NewStateLogger(eventCh <-chan Event) *EventLogger {
 					sw.matchCh <- event
 				}
 			}
-			DefaultLogger.Printf("    event received: %v\n", event)
+			slog.Info("event received", "event", event)
 			el.events = append(el.events, event)
 			el.lock.Unlock()
 		}