better support for pubsub / batch workers

[sanposhiho]

pubsub/batch workers don't receive the traffic always. They're busy sometimes and free for other periods.
Given Tortoise generates the recommendation based on VPA's recommendation and VPA takes a recent usage with more weight, it can suggest something inappropriate for those workers.
For now, the only way is to define minAllocatedResource in those tortoises so that tortoise doesn't scale down very much.

This issue is to make Tortoise generate better recommendation for them too.

---

Task list

• define maxReplicas / maxAllocatedResource in Tortoise API to limit scaling per workload #293