Too many unnecessarily long timeouts on known errors

[drastawi]

As a result of #454 running terraform on a project with an error in the config or one that contains a resource that had been manually deleted became extremely slow. The rules for retry are not centralized and vary widely from resource type to resource type.

There is a lot of value in retrying for a long time on 429 but for example, we should not wait 5 min to learn we made a typo or are missing field and get a 400. Sometimes it might be valuable to avoid a race condition for a 404, but that should not take 5min, maybe 30s.