[Bug] Prevent hot-reload of workflow definitions from disk

A Python worker reloads the workflow definition from disk every time a new workflow run is started, i.e. without requiring the worker process to be shutdown and restarted. So for example, a long-running workflow that CANs will "hot reload" like this on every CAN, and a non-sticky worker will hot-reload on every WFT.

This is a side-effect of implementation (worker sandboxing involves disabling import cache which results in re-reading from disk) rather than being desired behavior, and isn't appropriate in production worker deployments since users are likely to expect that during deployments they can overwrite files without any modifications to running processes until process restart.

So, let's find a good way to prevent the behavior. For example, one possibility might involve permitting the re-read-from-disk to continue, but having the worker copy the relevant files somewhere private, and importing workflows from there.

Es werden keine Dateien oder Tests genannt. Beginne damit, das im Issue beschriebene Sandboxing des Python-Workers und das Verhalten des Import-Caches nachzuverfolgen, und untersuche anschließend, wie Workflow-Definitionen über wiederholte Ausführungen hinweg geladen werden. Erledigt ist die Aufgabe, wenn ein laufender Worker seinen Workflow-Code unverändert beibehält, bis der Worker-Prozess neu gestartet wird, einschließlich bei langlebigen und nicht sticky Workern.