Currently we can queue up to 2 batches of 100 logs, but our specs have been updated with the following
SDKs MUST also have a hard limit of 1000 log events queued up to avoid causing your customers' applications going out of memory
We could extend our existing two-buffer-approach to a 10-buffer-ring, where on flush we 'walk' through all pending buffers (and we'd need to move our active-buffer-index update outside of the flush step)