l2assemble enhancements

[dopplershift]

• Handle re-issued volumes better
  • The loading from S3 problem
  • Validate the fix deployed to production is the right one
• Add logging for that case
• Start tracking latency of volumes (look at difference between current time and time of first chunk)
• Gracefully handle S3 write failure?
  • Maybe cache volume to disk?