Fix socket descriptor checks on Windows

[olszomal]

The socket range checks introduced in e98c735 break non-blocking socket operations on Windows. This causes random WSAENOTCONN errors in subsequent read/write attempts.

1. In the implementation of the socket() function in Windows, the nfds parameter is ignored and included only for compatibility with Berkeley sockets. On Windows, the FD_SETSIZE value contains the maximum total number of sockets in fd_set rather than the limit for for FD_SET() and FD_CLR() macros.
   https://learn.microsoft.com/en-us/windows/win32/api/winsock2/nf-winsock2-select

2. Also, the Winsock invalid SOCKET value is INVALID_SOCKET, and not any integer < 0 as on Unix.
   https://learn.microsoft.com/en-us/windows/win32/winsock/socket-data-type-2

My PR fixes checking Windows socket value according to Microsoft documentation.

[olszomal]

I could not reproduce the testing framework failure.

$ perl configdata.pm --dump
    /usr/bin/perl ./Configure --banner=Configured --debug -O1 -fsanitize=memory -DOSSL_SANITIZE_MEMORY -fno-optimize-sibling-calls enable-rc5 enable-md2 enable-ec_nistp_64_gcc_128
(...)

$ make test TESTS="test_tls13messages"
(...)
70-test_tls13messages.t .. ok    
All tests successful.
Files=1, Tests=17,  3 wallclock secs ( 0.23 usr  0.00 sys +  1.90 cusr  0.50 csys =  2.63 CPU)
Result: PASS

Could it be an issue with the testing framework itself?

[openssl-machine]

This pull request is ready to merge

[t8m]

Merged to all the active branches. Thank you for your contribution.