Problem: adapt, clarify docs ZMQ_ROUTER_MANDATORY

Solution:

* Document the new behaviour when generating 'ZMQ_POLLOUT' events
  for ZMQ_ROUTER sockets with 'ZMQ_ROUTER_MANDATORY' set to `1`
* Add clarifications for 'ZMQ_ROUTER' socket when
  'ZMQ_ROUTER_MANDATORY' is set to `1`

Fix ROUTER's xhas_out() in MANDATORY mode

Before this commit, xhas_out() was returning true regardless. This
was correct before the ZMQ_ROUTER_MANDATORY flag as introduced.
However, ZMQ_POLLOUT.

With this commit, _if_ ZMQ_ROUTER_MANDATORY is set, xhas_out() will
return false if ALL peer's outgoing pipes are full.

There is an outstanding high-level design question:

If ZMQ_ROUTER_MANDATORY is set, and zmq_poll() waits for ZMQ_POLLOUT
events, zmq_poll() will immediately wake up if only 1 pipe has
room to send, regardless of the peer, creating a busy loop of
zmq_poll() wake-up, zmq_send() (EAGAIN). There is no way for
the application to selectively wait for ZMQ_POLLOUT for specific
peer(s), which seems somehow necessary in ZMQ_ROUTER_MANDATORY.

This discussion will be addressed in a separate issue.
Signed-off-by: Marc Sune <marc@voltanet.io>
Signed-off-by: Fredi Raspall <fredi@voltanet.io>

Problem: CURVE server (connect) fails when client rebinds

Solution: if a CURVE server is using zmq_connect, the same session
will be used for any client "reconnect" (actual binds). This is
acceptable, so do not assert if zap_pipe already exists during the
handshake, but simply reuse it.
Fixes #2608

RELICENSE: Matt Connolly

Refs #2376

Problem: When using print_backtrace() on Linux with libunwind, printout of stack traces from multiple threads are interleaved. Solution: added static mutex to serialize printing of stack traces.

Problem: intermittent memory leak for req/rep send/recv. #2602 Solution: memory leak fixed.

Problem: When using print_backtrace() on Linux with libunwind, printout of stack traces from multiple threads are interleaved. Solution: added static mutex to serialize printing of stack traces.

Update zmq.txt

Fixed minor typo.

Problem: XPUB_MANUAL subscriptions not removed on peer term

Solution: remove the pipe from the real trie when a peer disconnects.
Also add a unit test that exercises the behaviour by reconnecting
a different socket and sending a message that matches.
Fixes #2601 and introduced by #2042

Solution: fix unit test to bind after setting the option.

Problems: ZAP can be set up incorrectly, CURVE and GSSAPI can be used only with ZAP

Problem: cannot cross-compile with Mingw-w64

Solution: we can use 'CMAKE_SYSTEM_VERSION' instead of
'${CMAKE_SYSTEM_VERSION}' for the 'if' clauses.

CMake fails to evaluate condition when CMAKE_SYSTEM_VERSION is
empty, which can happen with a default installation of Mingw-w64
in Linux.

Solution: use lower case ws2_32.lib to enable cross compilation
from platforms with case-sensitive filesystems.

When cross compiling the tests with Mingw-w64, CMake cannot
locate Ws2_32.lib

Solution: do not fail if ZAP is not enabled.
GSSAPI already provides authentication and can be used separately,
so it is a valid use case.

Solution: revert change that made ZAP mandatory.
The "Stonehouse" pattern, where CURVE is used only for encryption and
without authentication, is a valid use case so we should still
support it.
Also restore CURVE testing in the test_heartbeat.

Fixes #2594

Solution: if inproc://zeromq.zap.01 exists, which means ZAP is
enabled, abort immediately if it cannot be used (eg: out of memory)
or it is configured incorrectly (eg: wrong socket type).
Otherwise authentication failures will simply be ignored and
unauthorised peers will be allowed to slip in.

fix bug: #2592 dish client does not resend subscriptions to radio server after radio server restart

problem: for zmq radio/dish pattern, if the radio process restarts, the dish will not resend subscriptions to radio. And the result is that the dish will never receive any more messages.

solution: in session_base_t::reconnect (), take ZMQ_DISH into consideration when invoking hiccup method.

fix bug: coredump if set linger and immediate together

In function session_base_t::reconnect, if we set immediate to 1 and set linger, we will get into first block of reconnect function, and set pipe to NULL, but we forget to cancel timer of linger. Once timer tiggered, we will get coredump. Solution: cancel timer in the end of set pipe to NULL

RELICENSE: Ilya Kulakov

Refs #2376

Suppress linker warning 4221 for MSVC

Some #define switches cause the body of entire files to be omitted. This
causes a linker warning on Visual Studio 2017, for example

    warning LNK4221: This object file does not define any previously
    undefined public symbols, so it will not be used by any link
    operation that consumes this library

Since this is warning us about something that shouldn't be
earth-shattering news, we add a linker flag to suppress this warning on
MSVC builds.

Problem: use-after-free in test_poller

Solution: remove server socket from poller before closing it
Fixes #2581

Use OBJECT_DEPENDS and OBJECT_OUTPUTS for precompiled.hpp

Problem: abort at socket creation on Android with jzmq

Solution: don't set thread name on Android

Setting a thread name on Android may fail with "permission
denied" error and abort the process due to failed assertion.
Tested on Android 5 and 6 (two phones).
Strangely enough, it only happens on signed APKs and is fine
in debug. Using JeroMQ is not an option as we need TCP keepalive
settings and authentication which JeroMQ doesn't support.

Problem: REP leaves label msgs for dead REQ in pipe

Solution: roll back the pipe if writing messages other than the
first fails in router::xsend. Roll it back also when the pipe is
terminating.
Also add test case that reproduces the memory leak when ran with
valgrind.
Fixes #2567