Fix an msleep() that can use different mutexes to avoid assert panic

[rmacklem]

Summary

During recent testing as part of an IETF NFSv4 testing event,
I got an assert panic in sleepq_add(), due to a call with a
different mutex for the same wchan.

This turned out to be harmless, since the msleep() is on
a fake wchan that never gets a wakeup() and is simply a
time delay sleep.

To avoid the assert panic, this patch replaces the msleep()
with a tsleep().

Test Plan

With this patch, the test runs no longer assert panic'd.

[kib]

Can you use pause(9) then? You would avoid fake_wchan.

[rmacklem]

Replace tsleep() with pause() as recommended by kib@.