ZMQ fails with "Interrupted system call" when a signal is delivered to the process

[yurivict]

An application fails when it is waiting for a ZMQ message and a SIGALRM signal arrives.

The reason: errno=EINTR isn't properly handled. The system call poll() fails with errno=EINTR and it is being reported as an error, while the correct way to handle it is to re-run this call in such case.

Testcase:
long-zmq-poll-sleep.cpp

The above testcase waits forever until SIGALRM is sent to it from outside.

Once the signal is delivered it fails with this log:

$ ./long-zmq-poll-sleep 
SIGALRM handler installed
Initializing ZMQ context...
Creating PULL socket...
Binding to tcp://127.0.0.1:5555...
Socket bound. Calling recv() - this will sleep in poll() for a long time...
Waiting for messages (press Ctrl+C to interrupt, or send SIGALRM to test signal handling)...
PID: 55792

SIGALRM received (count: 1)
ZMQ recv() threw an exception: Interrupted system call (errno: 4)
Total SIGALRM signals received: 1

All system calls made by libzmq should be guarded and re-run when they fail with errno=EINTR.
This shouldn't be a responsibility of the caller of ZMQ functions to do this because every system call can fail like this.
libzmq itself should guard system calls.

[yurivict]

#4838