perf: set num_maskmem=3 for all inference paths

Reduces SAM2 temporal memory bank from 7 to 3 frames on both the
sync /detect endpoint and the async background job path to lower
GPU memory usage with minimal tracking quality impact.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

app.py

@@ -302,6 +302,7 @@ async def detect_endpoint(
                 output_path,
                 query_list,
                 segmenter_name=segmenter,
+                num_maskmem=3,
             )
         except ValueError as exc:
             logging.exception("Segmentation processing failed.")

jobs/background.py

@@ -40,6 +40,7 @@ async def process_video_async(job_id: str) -> None:
                 enable_gpt=job.enable_gpt,
                 mission_spec=job.mission_spec,
                 first_frame_gpt_results=job.first_frame_gpt_results,
+                num_maskmem=3,
             )
         else:
             detections_list = None