More robust use of our virtualenv

Prior to this change, the script tests if it's running inside a
virtualenv; and if it is, it assumes that it must be inside its own
virtualenv.

This change switches to testing for the activate binary in the
place we expect; and if it's found, using it directly. This avoids
false positives (running the script inside the wrong virtualenv) and
makes sure that we're running inside the right virtualenv.

algo

@@ -2,19 +2,17 @@
 
 set -e
 
-if [ -z ${VIRTUAL_ENV+x} ]
+ACTIVATE_SCRIPT="$( cd "$( dirname "${BASH_SOURCE[0]}" )" && pwd )/env/bin/activate"
+if [ -f "$ACTIVATE_SCRIPT" ]
 then
-  ACTIVATE_SCRIPT="$( cd "$( dirname "${BASH_SOURCE[0]}" )" && pwd )/env/bin/activate"
-  if [ -f "$ACTIVATE_SCRIPT" ]
-  then
     # shellcheck source=/dev/null
     source "$ACTIVATE_SCRIPT"
-  else
+else
     echo "$ACTIVATE_SCRIPT not found.  Did you follow documentation to install dependencies?"
     exit 1
-  fi
 fi
 
+
 case "$1" in
   update-users) PLAYBOOK=users.yml; ARGS=( "${@:2}"  -t update-users ) ;;
   *) PLAYBOOK=main.yml; ARGS=( "${@}" ) ;;