Mainline code locks `MethodCompileQueue_lock` in `ciEnv::register_method`. But with Leyden-s `UseGlobalCompileQueueLock = false`, we have separate locks, per compile queue. So it seems to stand to reason we should be locking those locks, not a global one.

I have not been able to attribute this gap to any bug or performance issue, but I think we better fix this proactively.