add a 1ms delay between re-subscribing to PDS instances (#1090)

Assuming 5k active PDS instances, this spreads things out over 5
seconds. had been seeing burst of errors when restarting relay, which I
suspect is related to too many sockets (and DNS lookups) at the same
time.

cmd/relay/relay/crawl.go

···
 import (
 	"context"
 	"fmt"
+	"time"
 
 	"github.com/bluesky-social/indigo/cmd/relay/relay/models"
 )
···
 		if err != nil {
 			logger.Warn("failed to re-subscribe to host", "err", err)
 		}
+		// sleep for a very short period, so we don't open tons of sockets at the same time
+		time.Sleep(1 * time.Millisecond)
 	}
 	return nil
 }