Description of problem:
It's possible for multipathd to deadlock if the marginal_path_* shaky path detection is enabled. This happens because the marginal path io stats thread and
the path checker thread grab locks in different orders.  It is also possible for
the marginal path io stats thread to cause multipathd to crash on shutdown 

Version-Release number of selected component (if applicable):
device-mapper-multipath-0.8.4-7.el8

How reproducible:
Very hard. I've seen it once.

Steps to Reproduce:
1. Configure marginal_path_* shaky path detection

defaults {
        marginal_path_double_failed_time 60
        marginal_path_err_sample_time 120
        marginal_path_err_rate_threshold 1
        marginal_path_err_recheck_gap_time 10
}

2. fail and restore a multipath device twice, within one minute

# echo offline > /sys/block/<device>/device/state

wait till multipathd notces the path has failed

# echo running > /sys/block/<device>/device/state

wait till multipathd has restored the path

# echo offline > /sys/block/<device>/device/state

wait till multipathd has restored the path
# echo running > /sys/block/<device>/device/state

3. Continue step 2 on multiple different devices, hoping that the checker loop will try to add the device to the list of shaky devices at the same time as a device is being removed.

Actual results:
multipathd deadlocks, no longer checks paths, and becomes unresponsive

Expected results:
multipathd does not deadlock.

Additional info:
The shutdown crash was found through code inspection while looking into this issue. I'm not sure if it even can be produced in a real world scenario.