[llvm] [ADT][ConcurrentHashTable] Refactor ConcurrentHashTable. (PR #71932)

[via llvm-commits]

avl-llvm wrote:

> Can we use llvm's `RWMutex` instead of `shared_mutex`. On macOS this uses `pthread_rwlock` which is a lot faster than the implementation in the stdlib (#70151).

Actually std::shared_mutex does not work well on unix also. I wrote custom locker for DWARFLinkerParallel. std::shared_mutex is used in this patch as common default. Will replace it with RWMutex to have a better default.

https://github.com/llvm/llvm-project/pull/71932