Batch processor sorting logic

[jmacd]

I have studied the current state of our batch processing logic. I think the original author of the code might find a different solution, but I couldn't pick this up without starting over. I found it had some undesirable behavior, particularly around single payloads that exceeded the maximum size; the organization of batching into two discrete phases with split() and concatenate() is IMO not a great approach.

I have started with a rewrite of the batch processor, starting with the first principles that I wish to see that break with the current version:

• Focus entirely on Logs to start
• Provide an option to sort the main payload by some dimension(s), particularly useful ones (in logs) are timestamp, trace ID, event name
• Maximum size is not optional, since we have u16 limits anyway
• Produce batches of the maximum size with max 1 residual small batch
• Deduplicate Resource and Scope attribute values

There aspects of the current code that I will keep:

• basic data type Vec<[Option; N]>
• select() helper function for iterating basic data type
• unify() logic for schema unifications

See also
#969
#347