md/raid5-ppl: Use atomic64_inc_return() in ppl_new_iounit()

Use atomic64_inc_return(&ref) instead of atomic64_add_return(1, &ref)
to use optimized implementation and ease register pressure around
the primitive for targets that implement optimized variant.

Signed-off-by: Uros Bizjak <ubizjak@gmail.com>
Cc: Song Liu <song@kernel.org>
Cc: Yu Kuai <yukuai3@huawei.com>
Reviewed-by: Yu Kuai <yukuai3@huawei.com>
Reviewed-by: Artur Paszkiewicz <artur.paszkiewicz@intel.com>
Link: https://lore.kernel.org/r/20241007084831.48067-1-ubizjak@gmail.com


Signed-off-by: Song Liu <song@kernel.org>