Tidy up htex worker pool start time recording (#3112)

The start time is needed as part of upcoming work on walltime-based worker
draining (so this PR makes it into an instance attribute)...

... and it can be made more accurate than the previous implementation by
measuring it earlier (most especially, prior to invoking the often expensive
probe_addresses call)

This PR rephrases a nearby log line which uses the word "starting" despite
being not in the start() procedure of the manager, to reduce ambiguity when
reading the code vs logs.

parsl/executors/high_throughput/process_worker_pool.py

@@ -140,7 +140,9 @@ def __init__(self, *,
             Path to the certificate directory.
         """
 
-        logger.info("Manager started")
+        logger.info("Manager initializing")
+
+        self._start_time = time.time()
 
         try:
             ix_address = probe_addresses(addresses.split(','), task_port, timeout=address_probe_timeout)
@@ -454,7 +456,6 @@ def start(self):
 
         TODO: Move task receiving to a thread
         """
-        start = time.time()
         self._kill_event = threading.Event()
         self._tasks_in_progress = self._mp_manager.dict()
 
@@ -504,7 +505,7 @@ def start(self):
         self.task_incoming.close()
         self.result_outgoing.close()
         self.zmq_context.term()
-        delta = time.time() - start
+        delta = time.time() - self._start_time
         logger.info("process_worker_pool ran for {} seconds".format(delta))
         return