Support multiple endpoints per socket

Sent messages will be scheduled among the endpoints according to socket
type: PUB will send the same data to all endpoints simultaneously, PUSH
will do round robin transfer.
Incoming data is fair queued between endpoints.

This is a feature of at least zeromq and nanomsg.

_____________
To use: in the device configuration, instead of specifying just one address,
specify a comma separated list e.g.

tcp://localhost:123,ipc:///tmp/socket

the connection method (bind/connect) applies to all endpoints in this case.
______________
Mixing binding and connecting endpoints is supported:
prefix "@" means "bind", "+" (or ">") means connect, e.g.

+tcp://localhost:123,@ipc:///tmp/socket,ipc:///tmp/asd

(in case of missing prefix, the default channel method is used for that
endpoint).

fairmq/FairMQSocket.cxx

@@ -11,3 +11,53 @@
  * @since 2012-12-05
  * @author D. Klein, A. Rybalchenko
  */
+
+#include "FairMQSocket.h"
+
+bool FairMQSocket::Attach(const std::string& config, bool serverish)
+{
+  if (config.empty())
+    return false;
+  if (config.size()<2)
+    return false;
+
+  const char* endpoints = config.c_str();
+
+  //  We hold each individual endpoint here
+  char endpoint [256];
+  while (*endpoints) {
+    const char *delimiter = strchr (endpoints, ',');
+    if (!delimiter)
+      delimiter = endpoints + strlen (endpoints);
+    if (delimiter - endpoints > 255)
+      return false;
+    memcpy (endpoint, endpoints, delimiter - endpoints);
+    endpoint [delimiter - endpoints] = 0;
+
+    bool rc;
+    if (endpoint [0] == '@') {
+      rc = Bind(endpoint + 1);
+    }
+    else if (endpoint [0] == '>' || endpoint [0] == '-' || endpoint [0] == '+' ) {
+      Connect(endpoint + 1);
+    }
+    else if (serverish) {
+      rc = Bind(endpoint);
+    }
+    else {
+      Connect(endpoint);
+    }
+
+    if (!rc) {
+      return false;
+    }
+
+    if (*delimiter == 0) {
+      break;
+    }
+
+    endpoints = delimiter + 1;
+  }
+
+  return true;
+}