Fix two issues that resulted in incorrect updateTypes

- Force snapshot updates to have updateType=new, this makes sense
  because they are new to Felix even if they're the result of updates
  for Typha.  Add machinery to the FVs to check.

- The new machinery found a flake, which I tracked down to
  Workiva/go-datastructures#180
  pin go-datastructures to pick up the fix:
  Workiva/go-datastructures#182.

glide.yaml

@@ -36,11 +36,10 @@ import:
 - package: github.com/mipearson/rfw
 - package: github.com/prometheus/client_golang
   version: ^0.9.0-pre1
-- package: github.com/gavv/monotime
 - package: github.com/onsi/ginkgo
   version: f40a49d81e5c12e90400620b6242fb29a8e7c9
 - package: github.com/Workiva/go-datastructures
-  version: ^1.0.39
+  version: af7475fb23e7561c87e7c1d17c12f46d732379bb
   subpackages:
   - trie/ctrie
 - package: golang.org/x/text

pkg/snapcache/cache.go

@@ -358,7 +358,12 @@ func (c *Cache) publishBreadcrumb() {
 				counterUpdatesSkipped.Inc()
 				continue
 			}
-			c.kvs.Insert(keyAsBytes, newUpd)
+			// Since the purpose of the snapshot is to hold the initial set of updates to send to Felix at start-of-day,
+			// all the updates that it stores should have type UpdateTypeKVNew since they're all new to Felix. Copy
+			// the KV and adjust it before storing it in the snapshot.
+			updToStore := newUpd
+			updToStore.UpdateType = api.UpdateTypeKVNew
+			c.kvs.Insert(keyAsBytes, updToStore)
 		}
 
 		// Record the update in the new Breadcrumb so that clients following the chain of