Restructure shutdown logic used during updates. (#19781)

- Don’t blindly trust that the service manager actually shut down the
  agent.
- Bump timeouts between each method of shutting down the agent to 30
  seconds to make it more likely that the agent actually shuts down
  before we try to shut it down a different way.
- Add logic to handle different possible locations for netdatacli pipe.

packaging/installer/functions.sh

@@ -819,35 +819,43 @@ stop_all_netdata() {
     if [ -n "${NETDATA_STOP_CMD}" ]; then
       if ${NETDATA_STOP_CMD}; then
         stop_success=1
-        sleep 5
       fi
     elif issystemd; then
       if systemctl stop netdata; then
         stop_success=1
-        sleep 5
       fi
     elif [ "${uname}" = "Darwin" ]; then
       if launchctl stop netdata; then
         stop_success=1
-        sleep 5
       fi
     elif [ "${uname}" = "FreeBSD" ]; then
       if /etc/rc.d/netdata stop; then
         stop_success=1
-        sleep 5
       fi
     else
       if service netdata stop; then
         stop_success=1
-        sleep 5
       fi
     fi
   fi
 
+  if [ "${stop_success}" = "1" ]; then
+    sleep 30
+
+    if [ -n "$(netdata_pids)" ]; then
+      stop_success=0
+    fi
+  fi
+
   if [ "$stop_success" = "0" ]; then
     if [ -n "$(netdata_pids)" ] && [ -n "$(command -v netdatacli)" ]; then
-      netdatacli shutdown-agent
-      sleep 20
+      for p in /tmp/netdata-ipc /run/netdata/netdata.pipe /var/run/netdata/netdata.pipe /tmp/netdata/netdata.pipe; do
+        if [ -f "${p}" ]; then
+          NETDATA_PIPENAME="${p}" netdatacli shutdown-agent && break
+        fi
+      done
+
+      sleep 30
     fi
 
     for p in $(netdata_pids); do