fix: create the RequestExecutingAgent ProcessPool in initialize

ProcessPool was built lazily on the first processPool() call, which happens
inside execute(), after getBulkRequests() has already loaded a whole
BulkRequest into memory. All MinProcess workers are forked at that instant,
and because a worker never unwinds the frame it was forked from, every one
of them pins that cycle's Requests for the lifetime of the agent.

A core dump of one worker from a 60-process pool running BulkRequest=3000
still held the parent's requestsToExecute list and getRequests dict:
3003 Request, 3764 Operation and 40973 File objects, 54.8 MiB of live data
in a process that only ever handles one request at a time.

Creating the pool at the end of initialize() forks from the main thread
before any request is fetched, and keeps the RequestDB engine out of the
workers as well.

Claude-Session: https://claude.ai/code/session_01Wj93RvpYkvmL2AGmUHneRN

Author: chrisburr

src/DIRAC/RequestManagementSystem/Agent/RequestExecutingAgent.py

@@ -115,7 +115,12 @@ def __init__(self, *args, **kwargs):
         self.__rmsMonitoring = False
 
     def processPool(self):
-        """facade for ProcessPool"""
+        """facade for ProcessPool
+
+        The pool is created by :meth:`initialize`. Creating it forks MaxProcess
+        workers, which inherit everything reachable at that point, so the first
+        call must stay outside of the execution cycle.
+        """
         if not self.__processPool:
             minProcess = max(1, self.__minProcess)
             maxProcess = max(self.__minProcess, self.__maxProcess)
@@ -273,6 +278,15 @@ def initialize(self):
         # # create request dict
         self.__requestCache = dict()
 
+        # # Spawn the ProcessPool here, while the agent is still small.
+        # # ProcessPool forks its workers, so whatever is reachable at that moment is
+        # # inherited by every one of them and, because the workers never unwind the
+        # # frame they were forked from, is pinned for the lifetime of the agent.
+        # # Creating the pool lazily from execute() meant forking in the middle of a
+        # # cycle, freezing a full BulkRequest worth of Requests into all MaxProcess
+        # # workers. Doing it here also keeps the RequestDB connection out of them.
+        self.processPool()
+
         self.__requestDB = RequestDB()
 
         return S_OK()