Add micro_block_raw_aio_drain_128_us and
micro_block_raw_aio_drain_256_us tests that submit N AIO writes
to a temporary file, wait for the eventfd signal, then time how
long it takes to drain all completions via next_completed_request().
This measures per completion syscall overhead and provides a
baseline before any batching optimizations.
Signed-off-by: Anatol Belski <anbelski@linux.microsoft.com>
