Block builder scheduler startup: learn state from worker updates

[seizethedave]

What this PR does

More progress on the block-builder-scheduler:

• Add a startup period where we're learning the state of the world. This includes:
  • committed offsets learned from Kafka
  • any job progress/completion updates we've learned from workers given block-builder-scheduler.startup-observe-time
• When the startup period if over, we compute the starting state from the observations and begin normal operation.
• Add a job assignment epoch mechanism to break ties when multiple workers report updates about the same job. Clients include the epoch in any job-related communications.

What's next?

• Flush updated offsets periodically back to Kafka.
• RPC + client layers.

Checklist

• Tests updated.
• Documentation added.
• CHANGELOG.md updated - the order of entries should be [CHANGE], [FEATURE], [ENHANCEMENT], [BUGFIX].
• about-versioning.md updated with experimental features.