replication: rewrite error handling + simplify state machines

* Remove explicity state machine code for all but replication.Replication
* Introduce explicit error types that satisfy interfaces which provide
  sufficient information for replication.Replication to make intelligent
  retry + queuing decisions

  * Temporary()
  * LocalToFS()

* Remove the queue and replace it with a simple array that we sort each
  time (yay no generics :( )

daemon/pruner/pruner.go

@@ -11,6 +11,7 @@ import (
 	"github.com/zrepl/zrepl/replication/pdu"
 	"github.com/zrepl/zrepl/util/envconst"
 	"github.com/zrepl/zrepl/util/watchdog"
+	"github.com/problame/go-streamrpc"
 	"net"
 	"sort"
 	"strings"
@@ -334,6 +335,14 @@ func (s snapshot) Replicated() bool { return s.replicated }
 
 func (s snapshot) Date() time.Time { return s.date }
 
+type Error interface {
+	error
+	Temporary() bool
+}
+
+var _ Error = net.Error(nil)
+var _ Error = streamrpc.Error(nil)
+
 func shouldRetry(e error) bool {
 	if neterr, ok := e.(net.Error); ok {
 		return neterr.Temporary()