selftests: ublk: make IO & device removal test more stressful

selftests: ublk: make IO & device removal test more stressful

__run_io_and_remove() is used in several stress tests for running heavy
IO vs. removing device meantime.

However, sequential `readwrite` is taken in the fio script, which isn't
correct, we should take random IO for saturating ublk device.

Also turns out '--num_jobs=4' isn't stressful enough, so change it to
'--num_jobs=$(nproc)'.

Finally we don't cover single queue test in `test_stress_02.sh`, so add
single queue test which can trigger request tag recycling easier.

With above change the issue in #1 can be reproduced reliably in stress_02.sh.

Link:https://lore.kernel.org/linux-block/mruqwpf4tqenkbtgezv5oxwq7ngyq24jzeyqy4ixzvivatbbxv@4oh2wzz4e6qn/ #1

Cc: Jared Holzman <jholzman@nvidia.com>
Cc: Shinichiro Kawasaki <shinichiro.kawasaki@wdc.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20250519031620.245749-1-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>