cpu affinity should be supported for the threads of io_context_pool and msgr-worker-*

Hi.
I try to deploy the ceph-nvmeof and do some performance tests. I find that setting the cpu affinity for the threads of io_context_pool and msgr-worker-* can contribute to much higher performance both for the aarch64/x86 machines .
Further more, for aarch64 machines, the performance result showes that deploying the reactor threads and the io_context_pool threads in the same ccl (core cluster, usually 4 cores/ccl) will get better performance that in different ccl. So, implementing this by default and give a parameter for choice in the ceph-nvmeof.conf may help users to get better performance.

No files or tests are named. Start by locating how io_context_pool and msgr-worker-* threads are created and how ceph-nvmeof.conf parameters are handled; compare the requested CPU placement across aarch64 and x86. Done should mean the affinity behavior is implemented with a configuration choice and performance results are checked.

Written by the indexing model from the issue text.