fix(image-batch): give each image its own agent instead of sharing one

ImageAgentBatchProcessor submits the same Agent object once per image and
never resets it, so image N is answered with images 1..N-1 still in context.
Memory grows for the whole batch and is re-sent on every later call, and
concurrent workers append to one short_memory at the same time.

Hand each submission a private copy with empty short-term memory. Copying
also closes the race, since no two workers touch the same object.

Where the agent cannot be copied -- a held thread lock, an HTTP connection
pool -- fall back to the original, matching AgentRearrange._clone_for_task.
The fallback returns the agent untouched rather than resetting it: resetting
an object another worker is mid-run on would be worse than the sharing it is
meant to fix.

Closes kyegomez#2046

Author: ayaangazali

swarms/structs/image_batch_processor.py

@@ -1,3 +1,4 @@
+import copy
 import os
 import time
 from concurrent.futures import ThreadPoolExecutor, as_completed
@@ -107,6 +108,14 @@ def _validate_image_path(
             )
         return path
 
+    def _isolate(self, agent: Agent) -> Agent:
+        try:
+            isolated = copy.deepcopy(agent)
+        except Exception:
+            return agent
+        isolated.short_memory = isolated.short_memory_init()
+        return isolated
+
     def _process_single_image(
         self,
         image_path: Path,
@@ -217,7 +226,10 @@ def run(
             for path in validated_paths:
                 for agent in self.agents:
                     future = executor.submit(
-                        self._process_single_image, path, tasks, agent
+                        self._process_single_image,
+                        path,
+                        tasks,
+                        self._isolate(agent),
                     )
                     future_to_path[future] = (path, agent)