Add Bench CMD for engines and replicas

[shuo-wu]

longhorn/longhorn#7200

[shuo-wu]

Some problems in this PR:

1. The seq-related bench may be meaningless. If multiple threads are enabled during a seq bench, seq means each thread is responsible for operating its chunk sequentially. But all threads issue IO requests simultaneously, is it actually a kind of rand IO? For example, when thread 1 operates offest 0 at this moment, thread 2 immediately takes the chance and operates offset 1MB before thread 1 tries to operate its next block (let's say offset 4096). This leads to a seq-IOPS-write benchmark result mismatching the fio result.
2. The IOPS-read related benchmark result looks weird. In my test env (single node, 8 cores, 32G memory), when the thread count is 32, the engine seq/rand-IOPS-read would reach the max (82000), which is close to the local path performance (seq/rand result is 80000/91000). If we increase or decrease the thread count, 24 or 48 for example, the IOPS result is down to a half (43000).
3. Not sure if libaio used by fio will lead to a difference between this cmd test result and the fio result

[mergify]

This pull request is now in conflicts. Could you fix it @shuo-wu? 🙏

[mergify]

This pull request is now in conflicts. Could you fix it @shuo-wu? 🙏

[mergify]

This pull request is now in conflict. Could you fix it @shuo-wu? 🙏

[mergify]

This pull request is now in conflict. Could you fix it @shuo-wu? 🙏