Set the memory limit to 200MB

Two reasons why this is OK:
- We have improved memory usage: GoogleCloudPlatform/google-fluentd#197
- This seems to be right for workloads that heavily use the page cache, but don't necessarily benefit from it: kubernetes/kubernetes#43916 (comment)

logging-agent.yaml

@@ -61,7 +61,7 @@ spec:
         resources:
           limits:
             cpu: "1"
-            memory: 500Mi
+            memory: 200Mi
           requests:
             cpu: 100m
             memory: 200Mi