Allow pod-disruption-budget for individual Sematic funcs

[augray]

Sometimes it's hard for people to implement checkpointing (ex: if they are relying primarily on a third-party lib that doesn't support it). People may in this case want to try extra hard to ensure that their job doesn't get evicted. One could associate a pod-disruption-budget (PDB) with such jobs. This is slightly risky, as it might force k8s to keep nodes around for longer than desired during e.g. upgrades. One possible mitigation to that would be to require that when users specify a PDB, they must also associate a pod-level timeout (see #822).