This PR fixes issue kevin1024#710 by properly closing the underlying socket. It
first uses `pool._put_conn` to keep the connection in the pool, and
later removes and closes it when the context manager exits.

I was unsure about the exact purpose of the `ConnectonRemove` class,
so I made minimal changes to minimize the risk of breaking the code
and there may be better solutions for fixing this issue.

For example, the `urllib3.connectionpool.HTTPConnectionPool` will
utilize a weakref to terminate pool connections. By appending our
connection to it, it will also take care of closing our connection. So
another solution could be to modify the `__exit__` in
`patch.ConnectionRemover` method and add our connection to the pool:

```py
class ConnectionRemover:
    ...

    def __exit__(self, *args):
        for pool, connections in self._connection_pool_to_connections.items():
            for connection in connections:
                if isinstance(connection, self._connection_class):
                    pool._put_conn(connection)
```

…-warning-2

This reverts commit fa789e9.

This reverts commit f414435.

…-warning-2

…ed StopIteration