Socket's safe handle is actually two layers of safe handles: SafeSocketHandle, which is the one exposed to the rest of the implementation (and publicly), and then InnerSocketCloseHandle, which actually owns the handle / file descriptor and which is wrapped by SafeSocketHandle.

This layering should no longer be necessary, adds a lot of complication to the implementation, and makes creating sockets more expensive (two SafeHandles for each instead of one).

We should look at merging InnerSocketCloseHandle back into SafeSocketHandle.