Develop a concurrent memory-limited batch processor

[jmacd]

Briefly: we have had trouble tuning the configuration of an OTel-Arrow pipeline so that data does not drop. The root cause of this is the exporterhelper configuration and concurrency behavior combined with the core batchprocessor configuration and concurrency behavior.

We will develop an alternative based on the the core batchprocessor by forking the code, and then:

1. Modify the batchprocessor to block requests until the batch response is received
2. Modify the batchprocessor to return error or no-errors to the producer
3. Modify the batchprocessor to send >1 request in parallel, whenever data is available, using background goroutines for each batch
4. Establish an in-flight-bytes limit that prevents more than N bytes from entering its internal queue, to limit the amount of concurrent work and overall resource usage by the rest of the pipeline.