feat: expose step param on /detect/async, default to 7

Step=7 matches num_maskmem and reduces TTFS by shrinking the first
SAM2 propagation segment from 60 frames to 7. Benchmark shows
negligible end-to-end overhead (YOLO detection is fast).

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

app.py

@@ -409,6 +409,7 @@ async def detect_async_endpoint(
     depth_scale: float = Form(25.0),
     enable_depth: bool = Form(False),
     enable_gpt: bool = Form(True),
+    step: int = Form(7),
 ):
     _ttfs_t0 = time.perf_counter()
 
@@ -533,6 +534,7 @@ async def detect_async_endpoint(
         mission_spec=mission_spec,
         mission_mode=mission_mode,
         first_frame_gpt_results=first_frame_gpt_results,
+        step=step,
         ttfs_t0=_ttfs_t0,
     )
     get_job_storage().create(job)

jobs/background.py

@@ -34,7 +34,7 @@ async def process_video_async(job_id: str) -> None:
                 segmenter_name=job.segmenter_name,
                 job_id=job_id,
                 stream_queue=stream_queue,
-                step=60,
+                step=job.step,
                 enable_gpt=job.enable_gpt,
                 mission_spec=job.mission_spec,
                 first_frame_gpt_results=job.first_frame_gpt_results,

jobs/models.py

@@ -38,4 +38,5 @@ class JobInfo:
     mission_spec: Optional[Any] = None  # utils.schemas.MissionSpecification
     mission_mode: str = "LEGACY"  # "MISSION" or "LEGACY"
     first_frame_gpt_results: Optional[Dict[str, Any]] = None  # Cached GPT results from process_first_frame
+    step: int = 7  # Segmentation keyframe step (matches num_maskmem)
     ttfs_t0: Optional[float] = None  # TTFS anchor: time.perf_counter() at endpoint entry