This change fixes issues affecting the socket instrumentation in general.

1)

When a call to socket() fails, there is no socket created,
but the performance schema instrumentation api mysql_socket_socket() did create
instrumentation for the (non existent) socket anyway.
This is a bug, that can:
- cause statistical errors in various socket summaries, as there is an extra socket counted
- cause leaks of instrumented sockets, because the extra socket created is not destroyed,
- cause the socket_instance table to become full in the long term,
  causing the socket instrumentation to totally fail after a while.

The fix is to only create instrumentation for a socket if creating the socket worked.

2)

When creating instrumentation for a socket, the code calls:
- init_socket(), to create instrumentation.
- set_socket_info(), to set some attributes.

Making both calls each time a socket is created is un necessary overhead.
A more efficient way is to pass all the needed parameters to init_socket(),
so that a single call creates an instrumented socket, with optional attributes if needed.

This is a performance improvement, affecting the same area as 1)

3)

When bind() fails for a socket, the instrumentation api mysql_socket_bind() did set
attributes on the socket used in the bind, regardless of whether the bind operation actually worked or not.
This can lead to a confusing instrumented state,
where the performance schema reports that a socket is bound to a specific address, even when such operation failed.

This is a minor bug, fixed in mysql_socket_bind().