atomic.cpp: Pool of spinlocks for shared_ptr instead of just single one

Pass pointer value of control block of shared_ptr and select an instance of spin lock based on pointer bits. Spinlocks can reside in a static array, they should be std::hardware_destructive_interference_size away.

Targets scenario of std::vector<std::shared_ptr<T>>> parallel copying. After contention for some lock, threads would by copying different pieces in case of fix is there. Currently I believe one would block others until it has done copying (due to the "unfairness").

This is not the same as #370 , these are separate issues, they address different scenarios, and each could be fixed separately from the other.

This one is ABI breaking as currently _Lock_shared_ptr_spin_lock() / _Unlock_shared_ptr_spin_lock() do not take parameters.

vNext note: Resolving this issue will require breaking binary compatibility. We won't be able to accept pull requests for this issue until the vNext branch is available. See #169 for more information.

Start with the _Lock_shared_ptr_spin_lock() and _Unlock_shared_ptr_spin_lock() entry points and review the vNext compatibility note linked from the issue. The work is complete when shared_ptr control-block pointer bits select among cache-line-separated spinlocks and the associated ABI changes are addressed on the vNext branch.