Fixes regression by commit 521e573b

Fixes regression by commit 7901b04c

Similar to commits 1686f4e8 and
30a5dd26

Fixes the !HAVE_THREADED_EVENT_LOOP build.

Not for MPD, but for other applications which might want to copy its
event loop, but do not need multi-threading.

Allows forward declaration.

Instead of using this as a base class implementing a virtual method,
the new class IdleEvent can be used as a variable, decoupling
IdleMonitor's internal state from the derived class.

This is similar to commit 30a5dd26
which refactored TimeoutMonitor to TimerEvent.

By making each SocketMonitor keep track of its ready flags, this
removes a lot of overhead from EventLoop::RemoveFD().

By bit-wise ANDing the reported flags with GetScheduledFlags(),
ERROR/HANGUP always get cleared.  This means the MPD event loop could
never report those conditions.

This just happened to break the Windows build because of the
`GetObject` macro in `windows.h`, so I added a kludge to
PollResultGeneric.hxx.

Don't attempt to initialize the io_uring subsystem more than once.

Relax the assertions.  This is necessary if BlockingCall() is used
before the thread is started.

This fixes a freeze bug in the NFS input/storage plugins: when libnfs
auto-reconnets after a failure, it installs the new socket on the same
file descriptor number.  MPD's attempt to unregister the old socket by
calling SocketMonitor::Steal() from NfsConnection::ScheduleSocket()
fails because the new/old socket number is not registered in epoll, so
epoll_ctl() returns ENOENT.  The problem is that it left
`scheduled_flags`, and so subsequent Schedule() calls will use
`EPOLL_CTL_MOD`, which will fail again and again.  Instead, we need to
use `EPOLL_CTL_ADD` to register the new socket.

Closes https://github.com/MusicPlayerDaemon/MPD/issues/806

Closes https://github.com/MusicPlayerDaemon/MPD/issues/756

These flags are output-only.  Using them here is misleading.

The event library uses various libnet.a classes,
e.g. SocketDescriptor.