threadshare: use bench to select appropriate Mutexes

Depending on the contention model and sync / async usage, several locking mechanisms are available:

• std::sync::Mutex.
• std::sync::RwLock.
• futures::lock::Mutex.
• parking_lot:*
• simple-mutex.

We should use benches to select the most efficient candidate.

Note that smol might be used as a basis for the executor and can lead to different results.

See !298 (comment 458654) and !282 (comment 448814).