Adding locks to lookup and cleaning ModuleEntry table and PackageEntry table to enable concurrent class unloading. These may not be cleaned during a safepoint in the future.

Verify that lookup under a lock doesn't hurt performance.

Also, it appears that stale ModuleEntry entries, that is, from classes that are unloaded, on the reads list are okay because the list is followed only to match with an existing ModuleEntry that is not from an unloaded class (so they should not match). PackageEntry might be the same (?).