Ensure that we flush chunks into permament storage once a full chunk is received

Currently, when using Live Traces and a full chunk is received we schedule async operation to flush the chunk from Redis to DB/OS.

Proposal

Introduce the following barriers into a process:

1. When a chunk is received, the persist operation is synchronous
2. When a final update is received, all chunks should be marked as persisted, and no longer be allowed to be appended

Outcome

This would improve a live traces resiliency, by:

1. Ensuring that all data is persisted
2. Ensuring that we can safely and timely archive all data without having Redis to expire them
3. Introducing barriers (flush/syncs) into the process

Related code:

    def schedule_to_persist
      return if data_persisted?

      Ci::BuildTraceChunkFlushWorker.perform_async(id)
    end

Patching

Additionally, we could consider accepting no more than 3-4 chunks per single PATCH, and return an accepted offset for Runner to resent. Runner, already today sends a chunked PATCHes, but this can be a way to ensure that other users adhere to this operation as well: limited patching operation.