refactor(compressors): consolidate pipeline and improve mask handling

mini-nav/configs/models.py

@@ -18,7 +18,12 @@ class ModelConfig(BaseModel):
     compression_dim: int = Field(
         default=512, gt=0, description="Output feature dimension"
     )
-    device: str = "auto"
+    device: str = Field(
+        default="auto",
+        description=(
+            "Device to use for model inference (e.g., 'cuda:1,3', 'auto', 'cpu')"
+        ),
+    )
     sam_model: str = Field(
         default="facebook/sam2.1-hiera-large",
         description="SAM model name from HuggingFace",