fix(ml): better model unloading (#3340)

* restart process on inactivity

* formatting

* always update `last_called`

* load models sequentially

* renamed variable, updated docs

* formatting

* made poll env name consistent with model ttl env

machine-learning/app/config.py

@@ -13,7 +13,8 @@ from .schemas import ModelType
 
 class Settings(BaseSettings):
     cache_folder: str = "/cache"
-    model_ttl: int = 0
+    model_ttl: int = 300
+    model_ttl_poll_s: int = 10
     host: str = "0.0.0.0"
     port: int = 3003
     workers: int = 1