Bug#37039422 Backport Bug#36682079: crash on thread_pool + audit_log with [SEMI]SYNCHRONOUS when killing connections

Problem: Using the audit log plugin together with the thread pool plugin could
result in SEGV.

Root cause: The audit log plugin would execute code when a connection is
killed, and this code could invoke the thd_pool_wait_begin() callback.
Since this was done from the waiting thread without attaching the connection
to worker thread tp_low_level_client_t::processing_thread would be nullptr.
This was not expected by thd_pool_wait_begin().

Solution: Unset THD::scheduler_data before calling end_connection(). That way
both thd_pool_wait_begin() and thd_pool_wait_end() return early and the
invalid tp_client_low_level_t::processing_thread is not dereferenced.
Additionally, specialize  error handling for poll and epoll when waiting for
events and allow a Windows-specific error mentioned in the bug report.

Change-Id: I7c2d2d83fd4e531d950895678f1ff4b07a066e89

Author: Dyre Tjeldvoll

mysql-test/include/check_windows.inc

@@ -0,0 +1,10 @@
+# ==== Purpose ====
+#
+# Check if we are executing on Windows.
+#
+# ==== Usage ====
+#
+# --source include/check_windows.inc
+# # Now check if $have_windows is 0 or 1
+
+--let $have_windows = `SELECT CONVERT(@@version_compile_os using latin1) IN ("Win32", "Win64", "Windows")`