Fix tracing memory leak (Heroic-api)

[lmuhlha]

Memory leak caused slow heroic request responses

Currently the thought is that a memory leak in the tracing library was causing this, only triggered by an increased number of metrics/tags

[lmuhlha]

• Running heroic with new JVM args and enabled tracing again
• Noticed the used memory on the gae node double in the time I had tracing enabled
• Logs don't look too out of the ordinary

[lmuhlha]

• Left this running on 2 machines to see if the issue could be replicated, but it didn't come up
• Might be a grpc issue and require more machines across regions running tracing to replicate this
• Going to wait on upstream issue resolutions to pick this back up again:
  Memory Settings - Memory settings census-instrumentation/opencensus-java#1767
  Decide what to do when async event queue is full - Decide what to do when async event queue is full. census-instrumentation/opencensus-specs#262
  Configure a max number spans in RunningSpanStoreImpl - Configure a max number spans in RunningSpanStoreImpl census-instrumentation/opencensus-java#1813

[dmichel1]

Fixed in https://github.com/census-instrumentation/opencensus-java/releases/tag/v0.22.0 and #501