fix netshutdown

[qianxichen233]

Current implementation of netshutdown_syscall has several issues:

1. shutdown() doesn’t close the file descriptor, even if how is specified as SHUT_RDWR. To close the file descriptor, we must additionally call close(). Our implementation would take the file descriptor if how is SHUT_RDWR, which leads to several other undesired behavior:

• call shutdown with SHUT_RD then SHUT_WR (or reverse) would not behave the same as SHUT_RDWR
• when the file desciptor is taken, all the associated resources havn't been cleaned (NET_METADATA, FS_METADATA)

2. Once the socket is fully shutdown (SHUT_RDWR), its local_addr would be set to None, and state will be set back to NOTCONNECTED. This gives an chance to reuse the file descriptor and bind it to a new address, which is not an allowed behavior in Linux.
3. Some state are not handled correctly. In Linux, shutdown would behave differently for socket in different state. For example, shutdown with SHUT_RD on socket that is in listen state will disconnect the socket directly, without the need to shutdown it with SHUT_WR again.
4. Connection state is not checked, if the socket is not connected, we shall return ENOTCONN error.
5. shutdown on AF_UNIX socket is not working. After a SHUT_RD on a UNIX domain stream socket, the peer application receives a SIGPIPE signal and the EPIPE error if it makes further attempts to write to the peer socket; After a SHUT_WR, Once the peer application has read all outstanding data, it will see end-of-file. Subsequent writes to the local socket yield the SIGPIPE signal and an EPIPE error. These behaviors are not matched.

For reference, below is how Linux does for shutdown:
In Linux, socket has two different kinds of state, the first type of state is similar to our ConnState, describing a high level of the state of the socket. However, it does not have state like CONNWRONLY or CONNWRONLY, as shutdown are handled in a competely different way in Linux. Another type of the state is about the current TCP state of the socket (TCP_CLOSE, TCP_SYN_SEND, etc). These two types of state works together to identify current state of the socket. However, shutdown in Linux does not rely on these state actually, the socket struct has a dedicated field called sk_shutdown, which is the place where shutdown state is stored. The sk_shutdown will be checked in send or recv to affect their behavior.